Author: Scott

The AI Knowledge Problem: Leibnitz, Wolfram, and LLMs

Gottfried Wilhelm Leibnitz (1646-1716), who famously invented calculus at the same time as Isaac Newton (Newton got the credit; we got the Leibnitz notation, thankfully), also believed that if we could discover the correct way of representing our world in symbols, we could calculate the world using logic and mathematics. He also invented binary mathematics and made significant advances in logic to this end.

Stephen Wolfram is on a similar mission to make the world computable. For about 40 years, he has been working diligently on this project, starting with Mathematica, then several years ago expanding that to the Wolfram Programming Language, and more recently to the Wolfram Physics Project. Along the way, Wolfram has expanded on the ideas of von Neumann, Turing, and Godel by describing the principle of computational equivalence, and by carefully differentiating between the traditional, formula-based methods of math and physics, and computational or algorithmic methods.

Large Language Models (LLMs) like GPT-4 (ChatGPT) provide text that sounds like their training data, given prompts that they can use to predict what text most likely comes next. They neither encodes things about the world based on knowledge of how the world works for possible later computation (Leibnitz), nor do they compute answers to questions about the world by feeding knowledge into a compute engine (Wolfram). They simply predict what they think a human would write based on about 1500-3000 words of written prompt.

What do these have to do with the AI Knowledge Problem?

They represent three fundamentally different ways of dealing with it.

Leibnitz, based on the belief of his time that the world was inherently knowable, started with quantifying knowledge about the world. He believed that if we could solve the problem of describing the world in a consistent, logical, and rigorous manner, we could then perform computation based on that knowledge to make new discoveries. Alas, Leibnitz was burdened with living in a time before computing machines were available for him to fully experiment with his ideas.

Wolfram, who is burdened with the additional knowledge of the past 300 years, most specifically that of Godel, who showed that there is a limit to what is knowable within any system and that nothing is 100% knowable, chose to start with computation. He seems to believe fundamentally that if we build a good enough compute system, then we can feed it knowledge and get good answers. He has had some success with this approach. Most of us use his technology, even if we don’t know it. Wolfram Alpha, his “search engine” built on Wolfram Language, powers big chunks of Siri and Bing.

LLMs basically ignore the knowledge problem altogether. The most generous thing that can be said about LLMs and knowledge is that they regress to the mean of the training data, and if the training data overwhelmingly supports one hypothesis, the LLM will likely reproduce that hypothesis and some supporting statements. Of course, it can be prompted in a way to argue against a prevailing hypothesis. Unless the OpenAI safety engineers have decided that doing so causes a problem.

So we have a few choices about how to approach the AI Knowledge Problem.

Option 1 – The Leibnitz Paradigm. We can start with knowledge, properly quantified and qualified, and add compute and a friendly interface. The challenges here are several, not the least of which is that there are fundamental limitations on knowability which will require us to qualify our knowledge with a measure of uncertainty. Luckily, we have developed the mathematics necessary to do this. There are societal barriers as well, especially that those who profit from certainty do not welcome uncertainty, and they certainly do not gladly accept falsification. But to advance our knowledge, we need both – to acknowledge uncertainty and to welcome falsification.

Option 2 – The Wolfram Paradigm. We can start with computation and add knowledge and a friendly interface. This has the advantage of appearing to be useful more quickly by providing answers earlier in the process of making the world computable. But we have a GIGO (Garbage In, Garbage Out) problem when we scrape the internet for the knowledge to feed our compute engine. Without a rigorous understanding of the quality and certainty of the information going into the compute engine, we cannot assess the quality of the outputs. The risk here is that we attribute rigor to the process based on the computation, but forget that the computation depends on the rigor of the data.

Option 3 – The ChatGPT Paradigm. We can ignore the knowledge problem and use AI and human intervention to ensure that the outputs of our tools conform to some vaguely-defined standard of acceptability. There are many problems here. First, what is the standard of acceptability? If we are building tools to solve problems, as OpenAI claims in their advertising, then we risk overlooking solutions to difficult problems because of an arbitrary standard of acceptability, which may very well rest upon the belief that “we already know the right answer.” Second, if OpenAI continues to refuse to make their tool Open, then the users will not be able to assess the rigor of the output at all. Third, and perhaps most fundamentally, LLMs regress to the mean of the training dataset. As LLMs get more capable, the training data is becoming something like “a weighted representation of the entire internet.” The predictable result of LLMs becoming authoritative sources is that we continue to divide into our increasingly polarized camps and throw experts at each other. Even if we escape the polarization, we are still regressing to the mean of what we already think we know, which is a terrible way of inventing new solutions to difficult problems.

Which Option do you think holds the most promise for helping us solve our most difficult problems?

The AI Knowledge Problem

Artificial Intelligence has a knowledge problem. The latest wave of big AI tools, Large Language Models (LLMs) like ChatGPT/GPT4 and Llama, don’t encode actual knowledge about the world. They are only models that predict what words should come next, based on many billions of words of training data. LLMs usually sound like they are providing “correct” answers to our prompts, but this is merely a side-effect of the fact that LLMs are trained on data that contains enough correct information that it usually ends up sounding correct.

We are already seeing people getting frustrated with LLMs for “hallucinating” or “making things up” when the LLM generates text that is wrong or nonsensical. In Europe, government regulators are hard at work trying to mitigate potentially harmful uses and outputs of LLMs. OpenAI, the company behind ChatGPT, is working hard to make sure their product is safe, and at the same time, they are contributing to the problem by selling ChatGPT as a problem-solving tool.

A Very Brief History of AI

In the 1940s and 1950s, mathematicians and computer pioneers (most of whom were mathematicians) got excited about the potential of machines to think like humans. In the first wave of AI, this looked primarily like programming computers to play chess and checkers. These programs are search algorithms – they build a tree of possible moves and then search it for winning moves based on rules that assign values to different board positions. It soon became apparent that because of the number of possible positions in chess (about 10^44), we wouldn’t have enough computing power to search deeply enough to beat humans at chess for a very long time, and researchers moved on to other problems.

The second wave of AI in the 1960s-1980s consisted largely of heuristics engines and expert systems. Heuristics engines use a set of shortcuts or rules of thumb to quickly find a “good enough” solution to a complex problem that might not be solvable with traditional methods. Most anti-virus software uses a heuristics engine to identify suspicious behaviors. Expert systems encode knowledge about a specific domain of expertise in a rigorous way and use formal logic to solve problems within that domain. Examples include programs that can generate and verify mathematical proofs and programs for medical diagnosis. These systems were very highly developed by the early 1980s but fell out of favor for two main reasons: (1) they were extremely expensive to develop and maintain, and (2) they behaved unpredictably when faced with inputs that didn’t map directly to the underlying knowledge.

In the 1990s, computing power caught up to the chess problem enough to allow Deep Blue to beat Gary Kasparov, the top human player of the era. It was a landmark moment in AI, but looking back, probably said more about the weakness of human chess players than the strength of computers. We still can’t compute the full game tree of chess, or even exactly how many legal, reachable board positions there are.

By the early 2000s, large pattern recognition problems started to gain favor, as the combination of increasing computing power and improved neural network methods captured the imagination of researchers and programmers. Significant advances were made in techniques that would eventually prove fruitful in areas like voice recognition, handwriting recognition, facial and object recognition, and language translation. In 2012, a major neural network advance made image recognition practical, and neural networks have dominated the development of AI since then.

In the 2010s- present, the advent of compute-as-a-service enabled by cloud computing and the use of parallel processing GPUs to implement neural networks have lowered the barriers to AI development. LLMs were initially made possible by this new compute environment, and are gradually being moved onto smaller platforms. What we haven’t seen yet is a return to the knowledge-based approaches of the 70s and 80s.

A Call to Action

As we are poised to become dependent on LLMs to enhance our productivity, it will be increasingly urgent that we build a knowledge base under these tools. If we fail to do so, we risk granting authority to LLMs that are fundamentally untethered from actual knowledge of the world.

In future posts, we will examine the human knowledge problem, ponder what a return to knowledge-based AI might look like, and consider how incorporating a bottom-up knowledge representation might help us solve some of the biggest problems in scientific research today.

How to Start Lifting Weights: A Guide for Gen X

I’ve recently had a few friends ask me for advice on how to back into the weight room after years or decades off, or how to get started as an over-40 person. I put together the first version of this guide for one of those friends, an extremely successful tech executive in his early 50s with some athletic background who has been out of the weight room for many years. For this post, I’ve updated and expanded on the original version to make it more accessible to a general audience.

Who this guide is for: This is a guide for people in their late 30s – late 50s who are either beginners in the weight room or who have taken a long time away from lifting and want to get back into lifting. It will work equally well for men and women. The goal is to build a balanced foundation of movement competence and strength to support daily life and self-sufficiency. If you follow this program consistently, you’ll become one of the fittest people you know, and you will be able to go hiking with the teenagers, chase the grandkids, help your parents and siblings, do your own yard work (if you like that!), and take great care of yourself and your loved ones. These are all great goals. The program is designed with longevity, safety, reliability, and consistency in mind. It is designed to be accessible to all levels of fitness and to novices and beginners.

What this guide isn’t. This isn’t a fully comprehensive guide on the topic; that would require a book-length treatment. I don’t spend a lot of time on exercise descriptions as there are decent videos available from reputable websites that cover how to safely and effectively perform the movements at levels from beginner to elite athlete. I also don’t provide a comprehensive list of exercises; there are literally hundreds of exercises and variations available. For a good comprehensive resource for exercises, Arnold’s Encyclopedia of Bodybuilding is an excellent place to start. Or jump on YouTube and start searching, but caveat emptor, it can be hard to separate the wheat from the chaff. (I will list several of the sources that I trust for lifting advice at the end of the post.) This also isn’t an intermediate, advanced, or elite program. It isn’t designed to build or test a 1-rep max on any exercise, and I recommend against doing that until one is more advanced.

This is a long post. There is a lot that can be learned in pursuing fitness, and I only scratch the surface in this post. It is organized in a way that I hope will make it easily accessible. Section 1 provides an overview of the basic movement patterns and lists of exercises for each pattern by experience level. Section 2 provides some general rules for how to approach learning and executing the exercises for both machines and free weights. Section 3 provides three example programs – one for two days/week of lifting, one for three days/week of lifting, and one for four days/week of lifting. Section 4 covers how to get started, how to progress into the routines, how to make progress week to week, and when to get help. Section 5 briefly covers how to integrate this program with cardio and interval (HIIT) training, nutrition and recovery, what to do when you get sore, and what to do if you have old injuries or develop new ones.

Section 1: Basic Movement Patterns and Exercise Selection

In the programs below, I don’t list specific exercises, but rather movement patterns. Everyone’s physical proportions are different, so some exercises work better/feel better/are less injurious for each of us. The key is to keep our strength balanced by apportioning the work across the relevant movement patterns. For our purposes, we’ll group exercises into seven movement patterns. Within each pattern, I list basic, intermediate, and advanced exercises. If you have some previous experience with lifting, go where you are comfortable. If you find that it’s harder than you remember, you can step back a level until you find your footing again. If you are a beginner at lifting, start with the Basic Level exercises and progress to intermediate and advanced versions as you gain strength and confidence.

1. Hip Hinge Movement Pattern: In the hip hinge, the back is neutral and static, the legs are neutral and static, and all movement is around the hip joint. It is the foundation of a proper deadlift and primarily engages the muscles of the hip (glute max and hamstrings), along with the spinal erectors in a static contraction. *The booty note: It has recently become quite popular, especially with women, to focus on developing the gluteus muscles. Although technically not strictly a hip hinge movement, I’ll include them here anyhow. If you want to round out your tush, add a glute movement to the routines below, but don’t drop the Hip Hinge movement from your routine.

Basic Level Hip Hinge Exercises:

  • 45-degree hip (back) extension: Video demo here. I’d rather you focus on your glutes than hamstrings on these, at least until your hamstrings build some strength.
  • Dumbbell Romanian Deadlift: Video demo here. Focus on pushing the hips back with a slight bend in the knees. Lower until hamstring flexibility limits movement. Focus on the glutes to raise back up. 
  • *Glute: various Glute machines, usually of the kick-back variety.

Intermediate Level Hip Hinge Exercises: 

  • Weighted 45-degree hip (back) extension. As above, but with a weight plate held at the chest or a plate/barbell held at arm’s length vertically below shoulders. Start adding weight when 20 unweighted is easy.
  • Deadlifts. Deadlifts are worthy of a long post all by themselves. At their simplest, they consist of placing a barbell on the floor directly in front of you, reaching down and grabbing it, and then standing up to a position of attention. If you have significant experience deadlifting with a coach, you should be safe on these. If they are new to you, find a good coach to help. 
  • *Glute: Hip thrust machines, or hip thrusts on the Smith machine.

Advanced Level Hip Hinge Exercises:

  • Romanian Deadlifts with a barbell. RDLs should be thought of as starting at the top of the movement, descending to the limit of hamstring mobility, then driving back up from the hips.
  • Stiff-Legged Deadlifts with a barbell. SLDLs are not the same as RDLs. SLDLs start on the floor, like a normal deadlift, but we use a stiff-legged (legs fixed at 10-20 degree bend in the knees) movement pattern (a hip hinge) to lift the weight.
  • Heavy Deadlifts.
  • Barbell good mornings. Same movement pattern as an RDL, but with the barbell across the back as in a back squat.
  • *Glute: barbell or dumbbell hip thrusts, wide stance squats.

2. Squat Movement Pattern: As in the hip hinge, the back is neutral and static, but now we squat down/back roughly as if sitting in a chair until the knees are past 90 degrees, then stand back up. There are many, many variants of the squat, here are a few of the more useful ones.

Basic Level Squat:

  • Machine leg press. Pin selector machines are generally ok. 45-degree leg press sleds are better. Pendulum squat machines are even better. Avoid any machines where you lie on your back and press straight up – too much unnecessary risk. Aim to ensure your range of motion gets the knees bent to greater than 90 degrees. It’s easy to ego lift on these with a shortened range of motion. This is the path to injury. On the other hand, for most people, bringing the knees all the way to the chest is too much range of motion, and shifts load to the lower back in a suboptimal position. 

Intermediate Level Squat:

  • Smith machine squat and variants. Back squats, front squats, sissy squats, feet forward back squats, etc. The most common errors here are not aligning your body to the machine, and not facing the correct direction for safety. For safety, if you lose control of the bar, make sure that it rotates in the direction that causes the safety catches to engage. Lots of people orient to the mirror instead of the safety. It’s loud, scary, and sometimes entertaining when they get stapled under the weight.
  • Goblet squats.
  • Bulgarian Split Squats with dumbbells. Most people only need to hold one dumbbell, and it should be in the hand on the same side as the working leg. These require some balance and practice. I recommend grabbing a piece of 1-inch PVC pipe or a broom handle (most gyms have these stashed in a corner somewhere) and using it for balance in the non-engaged hand.

Advanced Level Squat: 

  • Barbell squats, high bar or low bar
  • Barbell front squat
  • SS Bar/Yoke bar squats. These are especially useful if you have shoulder mobility limitations or elbow/wrist injuries that make it difficult or painful to back squat.

3. Vertical Pull Pattern: This pattern is simply pull-ups and variations. These can be done with a variety of grips, etc. to target the upper back and latissimus dorsi muscles (lats). Wider grips hit the rhomboids and teres major/minor more effectively, while narrower grips hit the lats more effectively. Pronated grips (palms away) put more stress on the back and less on the biceps; supinated grips (palms toward the face) shift some of the stress from the back to the biceps. Neutral grips (palms facing each other) are also available and can be useful if the other grips cause elbow issues.

*A note on pull-ups. These sometimes get a reputation as a big scary monster. The average American man can’t do one; and they tend to be a bit harder for women than for men since men tend to have more upper body muscle mass than women. I believe that anyone can build the strength to do pull-ups. Doing so may require progressing and building the components of the exercise, identifying and attacking weaknesses, possibly losing some weight, and some time and grit. But it’s like any other skill. Absent a limiting injury or illness, I believe anyone can get there. There are lots of good programs out there for getting your first pull-up. There are also some that are rubbish. A couple of general rules of thumb. Chin-ups are easier for most beginners than pull-ups. A shoulder-width grip (or narrower if comfortable) is easier for most beginners than a wide grip. If you can do a reasonably strict cable pull-down with about 80-85% of your body weight, you are probably strong enough to do a pull-up. Finally, pull-ups respond better than most exercises to training the eccentric movement, that is climbing or jumping to the top of the movement and then lowering yourself as slowly as possible.

Basic Level Vertical Pull:

  • Cable Lat pull-downs. 
  • Machine pull-downs.

Intermediate Level Vertical Pull:

  • Chin-ups
  • Pull-ups

Advanced Level Vertical Pull:

  • Weighted chin-ups
  • Weighted pull-ups

4. Horizontal Pull Pattern: Rows and variations. Can be done with a variety of grips, angles, and implements of destruction.

Basic Level Horizontal Pull:

  • Cable rows. In order of difficulty: narrow grip (easiest) – medium grip – wide grip (1.5x shoulder width). A pronated (overhand) grip is preferred for back development. A supinated (underhand) grip favors more bicep involvement at expense of back development.
  • Machine rows. Hammer Strength has a series of pull-down and row machines that are plate loaded and work pretty well. You’ll have to figure out how to align your body correctly to the machine. Caution: you’ll see a lot of ego lifting on these by guys doing one arm at a time and using a lot of body English. The fix is to do both arms simultaneously.

Intermediate Level Horizontal Pull:

  • Single-arm dumbbell rows. Minimize trunk twist involvement on these.
  • Chest-supported dumbbell or barbell rows.

Advanced Level Horizontal Pull: 

  • Barbell rows.
  • Pendlay rows.
  • Kroc rows.

5. Vertical Presses. In a vertical press, the line of force is from shoulder height to overhead, vertical relative to your spine. These are shoulder presses and variants.

Basic Level Vertical Presses:

  • Machine shoulder presses. These can be the pin-selector type machine, or the plate-loaded lever-type machines (such as Hammer Strength)

Intermediate Level Vertical Presses:

  • Dumbbell shoulder presses
  • Arnold presses
  • Barbell shoulder presses on the Smith Machine
  • Seated Barbell shoulder presses

Advanced Level Vertical Presses:

  • Standing barbell shoulder presses

6. Incline Presses. In an incline press, the line of force is between 30 and 60 degrees from vertical/horizontal. For most people, 30 degrees is recommended. Steeper inclines tend to present more risk of shoulder impingement. Some builds will favor a shallower incline of 10-20 degrees. Most gyms will have both fixed incline benches and adjustable incline benches.

Basic Level Incline Presses:

  • Machine incline presses. These can be the pin-selector type machine, or the plate-loaded lever-type machines (such as Hammer Strength)

Intermediate Level Incline Presses:

  • Dumbbell incline presses
  • Barbell incline presses on the Smith Machine

Advanced Level Incline Presses

  • Barbell incline press

7. Flat (Bench) Presses. In a flat press, the line of force is perpendicular to your spine.

Basic Level Flat Presses:

  • Machine Flat (Bench) Presses. These can be the pin-selector type machine, or the plate-loaded lever-type machines (such as Hammer Strength). 

Intermediate Level Flat Presses:

  • Dumbbell Bench Presses
  • Bench Presses on the Smith Machine

Advanced Level Flat Presses:

  • Barbell Bench Presses

Section 2: General Rules for Learning Your Way Around the Gym

If you are a beginner, or even if you used to lift but are just getting back into it, the array of equipment at the average gym can be overwhelming. Even on machines, where the movement pattern is defined to a large degree by the design of the machine itself, the diagrams aren’t usually very helpful and it can be hard to know how to set the machine up the right way for your unique size and proportions. This section gives you some rules of thumb for barbells, dumbbells, and machines that will help you find your way.

Single joint and compound movements.
For our purposes, we divide movements into either single-joint or compound. In a single joint movement, only one joint bends. In a compound movement, more than one joint bends. All of our movement patterns are compound movements, except for the hip hinge. The other single-joint movements in the programs below are isolation movements for the arms and legs, such as arm curls, triceps extensions, and leg curls.

Machine set-up for single-joint movements. For properly setting up machines, I see the most errors made in setting up the single-joint movement machines. The principle here is simple. The axis of rotation of the machine should be aligned as perfectly as possible with the axis of rotation of the joint involved. For a leg curl machine, the pivot point of the lever arm on the machine should be aligned with the pivot points of both of your knees. Once you find this alignment, adjust the rest of the machine to support the rest of your body in a way that keeps the joint perfectly aligned. Being out of alignment will make the movement feel awkward (you’ll likely feel a sticking point in the movement where it suddenly becomes more difficult), and over time it will create repetitive sheer stresses on the joint that will increase the probability of injury. It may be the case that you find a machine where you can’t get properly aligned because your proportions and the design of the machine don’t allow it. In this case, select a different movement for that muscle. In the case of cable/pulley machines for curls and triceps, it is generally best to have the elbows slightly forward of your body, the upper arms vertical, and to imagine fixing your elbows in space as you perform the movement.

Alignment for pressing movements: 
For vertical presses, you want a straight, vertical line between your elbow, wrist, and the load, and this line is perpendicular to the floor. This ensures that you are putting force in the opposite direction of gravity. It is ok, and generally desired, for the elbows to be a bit in front of the line between the tips of your shoulders. This is because your shoulder joint rides on the scapula, and your scapula are not parallel to the plane that runs through the tips of your shoulders. So we want the arms aligned with the scapula (this is called the scapular plane), which is generally about 15-20 degrees forward of the shoulders. In some cases, it is even preferred to have the elbows forward of the scapular plane (this is true in Arnold presses, and usually in barbell shoulder presses). For the Smith machine and barbell overhead pressing, you will have to get the bar around your head, which also necessitates the elbows being forward a bit. As you get the barbell over your head, you will want to align it more closely with the shoulder joints, or else it will fall forward.

For incline and flat presses, you want your elbows, wrists, and the load stacked vertically. For incline presses, this means that the bar (or DBs, or your hands) will “touch” between the nipple line and the clavicle. How far toward one or the other depends on the angle of the incline; if the bench is flatter (10-30 degrees), you will touch lower on the chest; if the incline is steeper, you will touch higher on the chest. For a flat bench press, the bar will touch at or a bit below the nipple line, depending on your proportions.

On machines, for incline and flat pressing variations, if the seat is too low, you’ll feel like your hands are getting jammed into your shoulders. If the seat is too high, horizontal presses start to resemble dips and incline presses start to resemble horizontal presses. For machine vertical (shoulder) presses, if the seat is too high, your hands will get jammed into your shoulders. If the seat is too low, you won’t be able to lower the weight to at or just above shoulder height.

Stability. There is a trend in gyms, popularized by glossy magazines, glitzy websites,  and clueless trainers, that we should do exercises in unstable positions. The theory is that this helps work “those little stabilizer muscles,” which has apparently become a physiological term of art. Do not do this when holding a weight. We want to be as stable as possible so that we can safely and effectively apply force to the load in order to make our bodies stronger. This is especially true when holding a weight over your head, face, or throat. In general, if you are lifting a weight and your feet are on the ground, you want to imagine grabbing the ground with your feet and evenly distributing pressure across the whole foot, then building tightness in the legs and hips, then building tightness in the core and upper back, then gripping the bar as if you are trying to choke the life out of it. After you have done this, you are ready to move the weight. This is a skill that takes some time and practice to learn and takes years to perfect. The earlier we start, the better. If you want to work on balance as a skill, the best way to do that is to go to a yoga class or an adult gymnastics class.

Section 3: Three Example Programs.

These programs are based on the time you have available for lifting. In Section 5, I’ll talk about recommendations for cardio and HIIT to support overall fitness and health and how to integrate those alongside the lifting.

2 Days/week:
If you can only fit 2 days a week of lifting into the schedule, I recommend a full-body program where you hit all of the major muscle groups each workout. The advantage is that you hit every muscle twice a week, which is generally good, but the workouts are a bit long in terms of total sets and time, so you may find yourself running out of steam toward the end. I find that adding a few shakes of salt to my water bottle helps with this. Do this routine twice a week on non-consecutive days.

Legs:
Hamstring curls: warm-ups, then 3 sets of 15-20
A Squat movement: 3 sets of 12-20
A Hip Hinge movement: 3 sets of 12-15

Back:
A Vertical Pulling movement: 3 sets of 8-15 (*pull-ups exception to reps)
A Horizontal Pulling (rowing) movement: 3 sets of 12-15

Chest/Shoulders:
A Vertical Pressing movement: 3 sets of 12-15
Option 1: An Incline (10-30 deg from flat) Pressing movement: 3 sets of 8-12
Option 2: A Flat Pressing movement: 2-3 sets of 8-20
*Especially in the beginning, you can do either incline or flat. Some people’s shoulders will tolerate one better than the other. As you get in better shape, you’ll probably want to do both.

Optional (but recommended) Arms:
Curl variation: 3-5 sets of 12-20, supersetted with
Triceps variation: 3-5 sets of 10-20

3 Days/week:
This is a traditional push/pull/legs split where you’ll hit everything once per week with a little more volume each time. For getting back into lifting, you may feel like you want to hit it more often, but you’ll make 85+% of the progress once a week that you will twice a week, and the workouts are a bit shorter in terms of time, which will allow you to generate a bit more intensity. It’s a nice balance. For this program, we add a couple of accessory movements for joint health and mobility as well. And we get to do arms twice a week. You can do M/W/F, T/Th/Sa, however it works for you. You can also simply program this on a 6-day “week” alternating a lifting day with a non-lifting day.

Day 1: Upper Body Push/Arms
A Vertical Pressing movement: 3 sets of 12-15
An Incline (0-30 deg from flat) Pressing movement: 3 sets of 8-12
A Flat Pressing movement: 2-3 sets of 8-20
A Shoulder Lateral Raise movement (dumbbells or cables): 2-3 sets of 20
Curl variation: 3-5 sets of 12-20, supersetted with
Triceps variation: 3-5 sets of 10-20

Day 2: Legs:
Hamstring curls: warm-ups, then 3 sets of 15-20
A Squat movement: 3 sets of 12-20
A Hip Hinge movement: 3 sets of 12-15
Calf Press movement: 2 sets of 20-25
Tibialis Raises: 3 sets of 20-25

Day 3: Upper Body Pull/Arms
A Vertical Pulling movement: 3 sets of 8-15 (*pull-ups exception to reps)
A Horizontal Pulling (rowing) movement: 3 sets of 12-15
Face pulls: accumulate 100 reps with a light weight.
Curl variation: 3-5 sets of 12-20, supersetted with
Triceps variation: 3-5 sets of 10-20

4 Days/week:
This is the same basic routine as the 3 days/week, but we add a second pull day. The reason for this is that a large number of shoulder issues are a downstream result of doing too much pressing relative to pulling, which creates imbalances in strength around the shoulder. From an evolutionary perspective, primate shoulder anatomy evolved primarily for swinging from trees, not for lying on our backs pressing heavy weights. The cure is to do more pulling than pressing. With the extra day, we get 2x the pulling, and some additional work for shoulder health (of course you can add it to the 3 Day program, if desired). Program on either a 7-day or 8-day week.

Day 1: Upper Body Pull/Arms
A Vertical Pulling movement: 3 sets of 8-15 (*pull-ups exception to reps)
A Horizontal Pulling (rowing) movement: 3 sets of 12-15
Curl variation: 3-5 sets of 12-20, supersetted with
Triceps variation: 3-5 sets of 10-20
Face pulls: accumulate 100 reps with a light weight. *see note below
Dead Hangs: 2-3 rounds of max time (aim to progress to 60 sec each)

Day 2: Legs
Hamstring curls: warm-ups, then 3 sets of 15-20
A Squat movement: 3 sets of 12-20
A hip hinge movement: 3 sets of 12-15
Calf Press movement: 2 sets of 20-25
Tibialis Raises: 3 sets of 20-25

Day 3: Upper Body Pull/Arms
A vertical pulling movement: 3 sets of 8-15 (*pull-ups exception to reps)
A horizontal pulling (rowing) movement: 3 sets of 12-15
Curl variation: 3-5 sets of 12-20, supersetted with
Triceps variation: 3-5 sets of 10-20
Face pulls: accumulate 100 reps with a light weight.
IYWTVs: 2 rounds of 30-sec each *see note below
Dead Hangs: 2-3 rounds of max time (aim to progress to 60 sec each) *see note below

Day 4: Upper Body Push
A Vertical Pressing movement: 3 sets of 12-15
An Incline (10-30 deg from flat) Pressing movement: 3 sets of 8-12
A Flat Pressing movement: 2-3 sets of 8-20
A Shoulder Lateral Raise movement: 2-3 sets of 20

Additional Exercises:
I included hamstrings specifically at the beginning of leg day for three reasons: a large number of men our age with “lower back issues” really have hamstring issues that cause the lower back to compensate. Having strong, mobile hamstrings helps this a lot. Second, similarly, a large number of knee issues are caused by hamstring weakness relative to quad strength. Third, getting the blood flowing to the hamstrings will improve the quality of movement and effectiveness of your hip hinge and squat movements and reduce the chance of injury on those. It’s a win-win-win. I prefer the seated leg curl machine as my primary hamstring movement and I prefer higher reps (15-20) and volume. I’ve torn a hamstring in the past and I don’t like loading these heavy for that reason. I’ve done as much as 10 sets of 10 reps, 3x/week for months on these with great success. 

Face Pulls: These are for shoulder health. The exercise is gaining popularity in gyms these days, and guys are loading up a ton of weight and making it into a weird bicep-heavy standing-row thing. This defeats the purpose and adds a ton of injury risk. Here’s a very good video of proper technique. You’ll notice that the rather big guy demonstrating these is using a very light weight. This is the way. I use 20-25 pounds and accumulate 100 reps over as few sets as possible. The first set is typically 40-50 reps. These are for rear delts and rotator cuff strength – i.e. for shoulder health, and also help with upper back strength and posture. Also, notice his grip – very important. For some reason I almost never see anyone using the proper grip on these. Any other way of holding the rope attachment should feel very awkward.

IYWTVs: These are rotator cuff rehab exercises I learned during my shoulder surgery recovery. When I neglect them, I feel it. Here’s a decent explanation video. I do these lying on a bench face down rather than on the floor. I prefer to do the IYWs with palms down and the Ts with thumbs down and frankly I’m not sure why she teaches thumbs up. It creates some extra external rotation stress that isn’t necessary or helpful. My physios taught palm down for all variants. I also do “Vs” which is arms at 45 deg between legs and perpendicular, palms up. I prefer to do 30-second holds with a very light weight. Start with no weight. I’ve never used more than 3 pounds per hand. That’s plenty hard. Cycle through all five, resting as necessary, 2-3 rounds. Important tip: focus on keeping your upper traps (shrug muscles) relaxed throughout. That will force the right muscles to activate.

Tibialis raises: I got these from Ben Patrick (@kneesovertoesguy) • Instagram photos and videos. I highly recommend working his level zero knees over toes into your routine. Tib raises and backward walking are the entry points.

Dead hangs: I got these from physical therapists Bob and Brad on the YT. The idea is to accumulate 3-5 minutes of time hanging from a pull-up bar. They go over how it improves shoulder health and how to progress into it here.

Section 4: How to Start and How to Make Progress

Starting and Progression: I recommend doing 1-2 working sets of each exercise for the first couple of weeks. Try different exercises and figure out what feels best to you. Learn what your strength levels are. Add one working set every couple of weeks as your work capacity increases until you get to the sets listed.

Rep Schemes: I’ve made this a relatively high-rep program. I find that generally works better for me for hypertrophy and fitness as I get older, and usually results in less wear and tear on the joints. Peaking for absolute strength is a totally different game, and I strongly recommend against doing anything less than 5 reps, even “just to find out” for the first year or two. The exception to this is pull-ups. If you already have pull-ups it’s fine to use them in your back day at fewer reps per set than listed. Progress toward accumulating the listed reps over more sets. For example, if you already have five good pull-ups, accumulate 15 reps in 5 sets of 3. Then add one or two reps each workout. So the next workout might be sets of 4-3-3-3-3, then 4-4-3-3-3, etc.

Progressive Overload Always Applies:  It will take you a few workouts to find what weight gets you to failure on each exercise within the listed window of reps. That’s fine. Once you know, for example, that you can do lat pulldowns with 100 pounds and fail at 15 reps, then add 5 or 10 pounds the next workout. You may get 12 or 13 reps. Over the next couple of weeks, work back up to 15. Add weight again. This is a basic linear progression. It’s enough for the first couple of years of training. If you get bored in the 12-15 range, add more weight and spend a couple of months in the 8-12 rep range. Programming here isn’t specific enough that it really matters much. But progressive overload (doing a little more each workout) matters almost 100%. You will eventually get to a point where you plateau. Keep doing the sets and reps. Learn how to increase intensity. The way to break the plateau as a beginner/intermediate is to continue to accumulate sets and reps, not to change a bunch of stuff. Patience and consistency win.

Boring is Effective: Resist the urge to change things. You need consistency far more than you need exercise variety. Picking one movement for each pattern is enough for at least a year. Trying to “mix it up” or “confuse the muscles” will hamper progress in most cases. The exception is that as you (re)gain comfort in the gym, you may move from basic movements to intermediate, to advanced ones. Let the excitement result from tracking your personal best on each exercise. If you previously managed 10 reps with 100 pounds, and this week you do 11 reps, that is a new personal record! Celebrate the progress.

Keep the main thing the main thing. I have a blog post about how to frame training generally. TL;DR: in order of priority: Consistency, recovery, technique, intensity, programming.

Get appropriate help. For the advanced exercises, if you don’t have experience with them already (and maybe if you do), get someone to coach you. I have a blog post about how to find a competent trainer. Shoot me a note or a video if you need advice on a particular issue. Or if you want, you can always fly me out for a week, and I’ll teach you everything you need to know!  And it will be fun.

Section 5: Cardio, HIIT, Nutrition, Recovery, and Injuries

Cardio and High-Intensity Interval Training (HIIT). I find that about 2 hours of zone two (HR ~120) incline walking per week, and 1-2 bouts of high-intensity intervals per week on a bike or rower make everything else better. For the intervals, I rotate around doing longer intervals (1:00 work/2:00 recovery, 5-8 rounds), medium intervals (:30 work/1:30 recovery, 5-8 rounds), and short intervals (:20/:10 work/recover, 8 rounds; :10/:20 work/recover, 8 rounds). You can do this on non-lifting days or on lifting days. If doing cardio on lifting days, it doesn’t much matter whether you do your Zone 2 work before or after lifting (I prefer after). If you do your HIIT training on a lifting day, do it at the end of your workout.

Nutrition. Talking about nutrition runs the risk of creating a religious war, but I’ll do it anyhow because it is important. In general, no matter what your goals are, more protein in the diet is better. If you reliably eat 1 gram of protein per pound of body weight per day, you will be amazed at how quickly you will get results. If your body fat percentage is above 20% for men or 30% for women, you can eat 1 gram of protein per pound of lean body mass (total body mass – fat mass). A popular objection/fear is that eating too much protein can interfere with kidney function. This has been debunked for people with healthy kidneys. The other frequent objection is that it’s hard to eat that much protein from real food. This can be true, especially if you aren’t used to it. But I find that it’s generally a good thing when I am both struggling to eat enough food and losing body fat. It beats the barely-losing-body-fat-while-struggling-not-to-gnaw-my-arm-off feeling of low-protein, low-calorie diets. Of course, there are many competing hypotheses in the nutrition world. I find I feel best eating my protein from real food sources. Save protein shakes for rare emergencies.

Recovery. I talked about recovery in some detail in my hierarchy of fitness priorities post.  In order of importance for recovery: sleep, nutrition, mobility work, hot/cold treatments, bodywork (massage, etc.). For many people in their mid-30s and older, especially men, sleep issues may be present and undiagnosed. I highly recommend getting a sleep study (they can be done at home now) if you know, or your partner tells you, that you snore and/or stop breathing at night.

On Injuries. If you have old injuries, talk to your doctor or, probably better, an experienced physical therapist and seek their advice on how you might modify your exercise selection or form to accommodate your particular set of battle scars. Luckily, being active again will likely reduce some of the stiffness from old dings and you will likely find that you regain some lost range of motion as well.

Most of us will experience some sort of pain as a result of being active. As they say, accumulating injuries is the price of not spending life sitting on the couch eating Cheetos. DOMS, or delayed-onset muscle soreness, is to be expected from any effective fitness routine. Its degree seems to vary greatly among individuals. I typically have some DOMS from every lifting workout, even with many, many years of experience. The best fix for DOMS seems to be movement and getting the blood flowing into the sore muscles. For this reason, it is not necessary to delay your next workout until the soreness has faded.

Occasionally, you may get hurt. I think of this in three categories: hurt, injured, and broken (with apologies to Dave Tate). Roughly, “hurt” is a minor tweak or strain that goes away in a few days to a week, or some mild tendonitis. I generally work through being “hurt” as motion and blood flow help the body to heal through the normal range of motion and reduce scar tissue accumulation. “Injured” is a muscle pull or tear or a joint issue that is bad enough that it can’t be worked through and requires some rest. Usually we can work around this injury and maintain most of our fitness while seeking some physical therapy for the injured area. “Broken” is worse – this is a surgery-level injury that requires major modifications to our training plan and may curtail it all together for a period of time. Hopefully, this won’t happen, but sometimes it does. The good news is that in a beginner program with moderate to high reps and moderate weight such as this one, it is unlikely to have something worse than a minor “hurt” occur.

Bonus Section: Fitness Myths for the Gen X Athlete

Myth 1: You can’t build muscle and strength if you are xx years or older.
Busted: This is a popular misconception, along with the belief that after about 40, the best you can hope for is to slow your decline. I disagree, and so does the research. It’s never too late to start. Building muscle and strength are extremely tightly correlated (and probably also causally related) to both longevity and health span (how long can we remain active and independent, roughly). My personal example is that in early 2023, at the age of 52, I reached a strength goal in the bench press that I had been pursuing for most of 10 years, and that was a lifetime personal record by almost 10%. In late 2022, I made a lifetime-best deadlift as well. And somewhere in between there, I made a new personal record for pull-ups, performing 8 more reps than I did when I was a Division I competitive swimmer, and weighing about 65 pounds more than I did in college. The best day to get started was yesterday, the second-best day is today.

Myth 2: Doing xx exercise is inherently dangerous.
Busted: This one is usually applied to exercises like squats and deadlifts, and also frequently to other compound barbell lifts. While all exercise carries some element of risk because it is inherent in the nature of reality that risk is never zero, most exercises can be performed in a way that the benefit greatly outweighs the risk. There may be specific injuries that limit you from performing specific exercises, but there will definitely be at least one exercise from each movement pattern that you can do safely and effectively.

Myth 3: This one is for the ladies. It goes like this: I’m worried that if I lift heavy weights, I’ll get “bulky.”
Busted: This is the first cousin of “I just want to get toned.” Strength training causes two main adaptations in muscle tissue: hypertrophy (growth of the muscle itself) and increases in strength (the ability of the muscle tissue to produce contractile force). The two are related. In most cases, hypertrophy also causes an increase in the ability to produce force. However, one can train in a way to increase strength with little to no hypertrophy. It is really hard to build a large amount of muscle mass (“get bulky”). For natural/unenhanced men (men who are not using anabolics), one can realistically only expect to gain about one pound of muscle per month after the first year of training (an untrained man may be able to gain about 20 pounds of lean mass in the first year). For unenhanced women, it’s probably about half of that. There are genetic factors that apply here. Some can gain a bit more; some will struggle to gain a bit less. The vast majority of women who appear “bulky” due to muscle mass are enhanced. With hard training and on-point nutrition, a natural woman may gain about 5 pounds of muscle mass in a year, and due to having much less testosterone than men or enhanced women, her upper limit of muscularity won’t result in a bulky look. If she also loses 5 pounds of fat mass, the change in her appearance toward athleticism and health will be obvious to everyone around her. And her confidence and mood will improve as well, both from the results and from pride in the discipline it took to achieve.

Myth 4: The exercise science says I should do x or y or some other thing.
Busted: I’m a huge fan of the scientific method. And I think there are some significant problems with how we do research these days. These are topics for another post. My two basic observations on peer-reviewed science and exercise are (1) you can find a research paper to support any crazy thing you want to do, so that’s nice and also not particularly useful, and (2) in sports science, practice tends to lead research by decades. That is, Usain Bolt’s coach knows how to train athletes to run fast, and researchers are frequently coming in behind the events to figure out the physiological mechanisms that cause the methods to work. Find what works for you and trust it. Modify if/when necessary. To fully evaluate a program takes about a year, so don’t change lots of variables quickly. As a beginner, follow a beginner programmer for at least a year before you change anything. After a year or two, you might know enough to try changing one thing. Be patient. It’s the long game that matters.

Conclusion.

Good luck getting back into the gym. This program will put you into the top 10% of fittest people our age very quickly, and if you want to progress to more advanced training it will have provided a solid foundation. As with anything in life, there is always more to learn, and this rather long blog post is far from comprehensive on the topic. Let me hear how it goes and send me any questions you might have.

In strength,
Scott

Resources:

For excellent video demonstrations of most of the exercises listed in this program, follow Alexander Juan Antonio Cortes (AJAC) on Instagram here. I’ve met AJAC, and I’ve followed his blog for years. Although we have some minor differences of opinion, he knows his stuff and I trust his advice. I also recommend his programs.

For excellent video demonstrations of some of the exercises and very good instructional breakdowns of the underlying movement principles, follow Juggernaut Training (Chad Wesley Smith and Marisa Inda). Chad and Marisa are both international elite powerlifters. They have a mix of foundational and advanced videos. Their series on effective core bracing is especially good.

EliteFTS.com (YouTube channel here) has hundreds of videos from a large variety of experts on all topics of lifting. Their material tends to be more tailored for advanced lifters, but they do have some good introductory material as well. It’s worth keeping in mind that most of their experts are enhanced. Recommended for those who are contemplating getting into more advanced lifting or competing in strength sports.

AI Safety: A Partial Critique of Russell’s “Provably Beneficial AI”

With all of the recent advancements in AI, especially the November release of ChatGPT and last week’s upgrade of the underlying GPT engine to version 4, there is a lot of ongoing discussion about AI Safety.

AI Safety is a term that has come to mean a lot of things, and as such, it requires some further elaboration so we know what we are actually talking about. According to Wikipedia (Mar 21, 2023): “AI safety is an interdisciplinary field concerned with preventing accidents, misuse, or other harmful consequences that could result from artificial intelligence (AI) systems. It encompasses machine ethics and AI alignment, which aim to make AI systems moral and beneficial, and AI safety encompasses technical problems including monitoring systems for risks and making them highly reliable. “

For this post, I’ll be having a conversation with ChatGPT 4 to provide an introduction to one concept that lives within AI Alignment. The concept is called “Provably Beneficial AI” and has been proposed, refined, and advocated primarily by Stuart Russell of the University of California at Berkeley, where he is the Director of the Center for Human-Compatible AI, a professor of Computer Science, and the Smith-Zadeh Chair in Engineering. He literally wrote the book on AI.

I did the interview twice, once with the GPT 3.5 engine and once with the GPT 4 engine. GPT 4 generally provided more detail in its responses, with a strong tendency to use numbered lists. GPT 3.5 was more conversational, but provided less detail. I used the same prompts for both, and since I did GPT 3.5 first, the GPT 4 prompts might feel a little clunky, but I wanted to be able to compare the two and get a feel for the differences. In the interview below, I am S: and ChatGPT-4 is GPT.

Enjoy!

S: Briefly explain Russell’s concept of “provably beneficial AI”

GPT: Russell’s concept of “provably beneficial AI” refers to the development of artificial intelligence systems that can be mathematically proven to act in ways that are beneficial to humans. This concept aims to ensure that AI systems consistently work towards the best interests of humanity, avoiding unintended negative consequences.

In order to achieve this, the AI must be designed with certain principles in mind, such as uncertainty about its objectives and deference to human preferences. By constructing AI systems that acknowledge their own uncertainty and learn from human input, we can work towards creating AI that remains aligned with human values and provides benefits without causing harm.

S: What is the strongest critique of Russell’s “provably beneficial AI?”

GPT: One of the strongest critiques of Russell’s “provably beneficial AI” concept is the challenge of defining and quantifying human values and preferences. Human values are complex, diverse, and often contradictory, making it difficult to create a universally accepted model that can guide AI systems in a provably beneficial way. Some specific concerns include:

  1. Value alignment: Aligning AI systems with human values is challenging because values can vary significantly between individuals, cultures, and over time. It is unclear how an AI system can be designed to account for these variations and adapt to changing preferences without risking unintended consequences.
  2. Incompleteness and ambiguity: Human preferences are often incomplete or ambiguous, which makes it hard for AI systems to make decisions that are guaranteed to be beneficial. A provably beneficial AI needs a way to handle such situations while remaining aligned with human values.
  3. Preference manipulation: A learning AI system designed to be deferential to human preferences may inadvertently encourage humans to express preferences that are not in their long-term best interests, leading to potential manipulation of the AI system for short-term gains.
  4. Moral dilemmas and ethical trade-offs: Real-world situations often involve moral dilemmas and ethical trade-offs that can’t be easily resolved. Designing a provably beneficial AI that can navigate these complex scenarios in a way that respects diverse perspectives and ethical considerations is a significant challenge.
  5. Technical difficulties: The development of provably beneficial AI requires significant advancements in AI safety research, including robust methods for ensuring AI systems learn and follow human values, even as they become more capable. This is a difficult technical problem that has not yet been fully solved.

S: Didn’t Russell specify a big data approach to defining human values and objectives by essentially looking at what decisions humans made historically and inferring the underlying values and objectives?

GPT: Yes, Stuart Russell did suggest an approach to defining human values and objectives by examining the decisions humans have made historically, with the idea that AI systems can learn these values and preferences through a big data analysis. This approach is part of the broader concept of “inverse reinforcement learning” (IRL), where AI systems learn to infer the underlying objectives or values that drive human decision-making by observing human behavior.

However, despite the promise of this big data approach, it is not without challenges and limitations. Some concerns include:

  1. Biases and inaccuracies: Historical data can be biased or incomplete, reflecting cultural, social, or political norms of a particular time period. These biases may not accurately represent the values and objectives of all humans, potentially leading to AI systems that reinforce existing inequalities or injustices.
  2. Changing values: Human values and preferences evolve over time, and what might have been considered acceptable in the past might not be considered acceptable today. AI systems need to be adaptable and able to recognize these shifts to remain beneficial.
  3. Overgeneralization: Inferring human values and objectives from historical data can lead to overgeneralization, where AI systems may not be able to distinguish between genuinely shared values and preferences that vary between individuals, cultures, or contexts.
  4. Incompleteness and ambiguity: As mentioned earlier, human preferences can be incomplete or ambiguous, making it difficult to infer clear, universally applicable objectives from historical data.
  5. Ethical dilemmas and trade-offs: Learning human values from historical data does not necessarily provide AI systems with a clear framework for resolving complex ethical dilemmas or making trade-offs between competing values.

Overall, while Russell’s big data approach to defining human values and objectives offers a potential path forward, it is not without significant challenges that need to be addressed in order to develop truly provably beneficial AI systems.

S: Isn’t there a more fundamental problem with this big data approach? Specifically, that there are many cases where humans were forces to choose a course of action from a set of possibilities, all of which were bad options and not aligned with our values. For example, it is hard to argue that Truman’s decision to use nuclear weapons against Japan was his “preference,” but neither was a mass invasion of the Japanese home islands. In fact, all really hard ethical decisions are between bad alternatives, so using a big data approach to categorize the decision that was made as a “preference” seems to be very problematic, if we want AI to avoid making the same types of mistakes as humans.

GPT: You raise a valid concern about the limitations of the big data approach to defining human values and objectives. Indeed, there are situations where humans face difficult ethical decisions and must choose between suboptimal options, none of which align perfectly with their values. In these cases, inferring preferences from historical data may not accurately represent human values and could lead to AI systems making similar mistakes or perpetuating undesirable outcomes.

This fundamental problem highlights the importance of considering additional factors when designing provably beneficial AI:

  1. Context sensitivity: AI systems should be able to understand the broader context of human decision-making and recognize that certain decisions may have been made under duress or constrained by limited options, rather than as a true reflection of human preferences.
  2. Ethical reasoning: AI systems should be equipped with robust ethical reasoning capabilities that allow them to navigate complex moral dilemmas, even when historical data may not provide clear guidance on the best course of action.
  3. Human guidance and collaboration: Developing AI systems that work closely with humans and actively seek their input in situations with ambiguous or conflicting preferences can help ensure that the AI’s actions remain aligned with human values.
  4. Ongoing learning and adaptation: AI systems should be able to learn from new information, societal shifts, and evolving human values, allowing them to update their understanding of human preferences and make better decisions over time.

In summary, while the big data approach to inferring human values and objectives has its limitations, it can still be a useful starting point for understanding human preferences. However, to develop truly provably beneficial AI, it is crucial to combine this approach with other techniques that address the inherent complexity and nuance of human decision-making and values.

S: Human guidance and collaboration seems to imply that moral values are relative to who we ask. Is Russell’s approach inherently one of moral relativism?

GPT: Stuart Russell’s approach to AI value alignment does not explicitly endorse moral relativism, but it does recognize the diversity and complexity of human values, preferences, and cultural norms. By emphasizing the importance of human guidance and collaboration, Russell’s approach aims to create AI systems that respect and adapt to this diversity, rather than adhering to a single, predefined set of moral absolutes.

In practice, this means that AI systems designed using Russell’s approach may exhibit a degree of moral relativism, as they attempt to align with the values of the humans they interact with or the societies they operate within. However, the goal is not to embrace moral relativism for its own sake, but rather to ensure that AI systems are beneficial and aligned with human values, which can be diverse and even contradictory.

It’s important to note that balancing between moral relativism and moral absolutism is a challenge when designing AI systems. To create provably beneficial AI, researchers and developers must grapple with the complexities of human values and ethics, and strike a balance that respects diverse perspectives while still upholding some level of universally accepted moral principles.

S: How might we decide what are universally accepted moral principles? Who should help decide that? It seems like the greatest philosophical minds of all time have been working on this problem since before Plato, and we haven’t reached any agreement yet.

GPT: You are correct that determining universally accepted moral principles is a complex and longstanding challenge, one that philosophers have been grappling with for centuries. While there may never be a complete consensus on universally accepted moral principles, there are some approaches to consider when determining which principles to incorporate into AI systems:

  1. Cross-cultural consensus: Analyzing moral principles and values that are widely shared across different cultures and societies can help identify common ethical themes that may serve as a basis for universally accepted principles.
  2. Interdisciplinary collaboration: Involving experts from various fields, including philosophy, ethics, sociology, psychology, and computer science, can help develop a more comprehensive understanding of moral principles and their applicability to AI systems.
  3. Public involvement and deliberation: Engaging a diverse range of stakeholders, including the general public, in discussions and deliberations about moral principles and AI ethics can help ensure that multiple perspectives are considered and that the resulting principles are more widely accepted.
  4. Iterative refinement: Recognizing that universally accepted moral principles may evolve over time, it is important to establish ongoing processes for revising and updating these principles as societal values and norms change.
  5. International frameworks and guidelines: Building on existing international ethical frameworks, guidelines, and conventions related to human rights, environmental sustainability, and other global concerns can provide a foundation for developing universally accepted moral principles for AI systems.

Ultimately, determining universally accepted moral principles for AI is a collaborative and ongoing process that requires input from a diverse range of stakeholders, from philosophers and ethicists to policymakers and the general public. While achieving perfect consensus may be impossible, striving for a shared understanding of core moral principles can help guide the development of AI systems that are more broadly beneficial and aligned with human values.

Overall, not a bad summary of the goals and some of the potential pitfalls of Russell’s approach.

A Funny Thing Is Happening with ChatGPT

The latest version of ChatGPT, (based on GPT-4) is about a week old, and already something funny is happening.

A Good Start

People are finding some pretty cool ways to use ChatGPT to help with their normal work.

One user gave ChatGPT a description of his company’s mission and a rough job description; GPT wrote a formal job description, and it was posted to a job board minutes later.

Another user had GPT interview him and write content based on the interview.

One friend had it draft a wealth management plan.

Another had it build the financial policy manual for his company, including lists of reports and deliverables, deadlines, and cost estimates.

People have figured out that they can get GPT to change its writing style by providing examples, and that they can even have it to write legal documents.

Several people have had it draft marketing emails and web site copy to sell their products.

For the first few days, it seemed the sky was the limit with this amazing new tool.

And Some Frustration

In the last couple of days, I’ve increasingly been hearing about problems. Users are getting frustrated with the tool misbehaving and doing unexpected things.

In one case, for a user who was having it create budgets and financial plans, it started changing hourly rates and budget lines on its own. When confronted, it apologized profusely for the mistake.

Other users have been frustrated that it seems to change its mind if you ask it the same question more than once.

One guy was puzzled when he asked it to cut and paste some text into a new response, but it wrote a new paragraph on the same topic instead.

And of course, some users have tricked it into doing nonsensical things – like writing a set of instructions for patching a hole in the wall of a house made entirely of cheese.

What’s Happening?

What’s happening here? Is GPT misbehaving? Making mistakes? Deliberately undermining its users?

Without getting into the technical explanation for each of the types of “errors” above (and there are relatively simple explanations based on how GPT works for each of them), I think there is a meta-error happening. And we, the users, are the ones making it.

We have been so impressed with our initial experiences that we are treating GPT as something like an experienced Executive Assistant and expecting that level of performance from it. It is performing well enough that when it misses the mark, we are getting frustrated with it in the same way we would if it were human.

The problem is, GPT isn’t built to be an Executive Assistant. It is built to be a language model. What that means is that it does one thing very well – it predicts what word should come next given the previous couple thousand words, based on a corpus of about 500 billion words of written language (for more on how it does that, see my previous blog post here.).

It’s worth repeating for emphasis: the only thing GPT does is generate text that looks convincing by modeling a statistical process to predict what word should come next.

What it doesn’t do is… everything else. It doesn’t build plans. It doesn’t apply logic. It doesn’t understand meaning. It doesn’t understand consequences. It doesn’t do rigorous research. It doesn’t formulate new ideas.

But GPT’s text generation model is good enough that it looks like it does those things. The training data includes many, many examples of those types of writing, so it produces output that is convincing and might even be reasonably accurate. But any accuracy or inaccuracy GPT provides is a side effect of what is in the training data. 

So, if you ask GPT for a financial plan, it generates text that looks like a financial plan. If you ask it to explain the logic of Zeno’s Paradox, it generates text that looks like an essay on that topic. If you ask it to explain what it means to be alive, it generates text that looks like an answer to that question. If you ask it to provide references on a certain topic, it generates text that looks like a list of references (some of them are decent references, and some of them are rubbish). 

What we are seeing is a disconnect between what we expect GPT to do and what it is actually capable of doing. This disconnect seems to be present in large part because GPT sounds so human that we expect it to behave like a human. GPT may be a victim of its own success. But it turns out that humans do a lot more than generate streams of words that sound human. We are expecting way too much from GPT. 

It is important to understand what GPT does, and what it doesn’t do. The reactions of some pretty smart people experimenting with the tool suggest that together, we (the users and OpenAI) haven’t done well at communicating the limitations of GPT. I hope we become more educated users soon – before we start treating GPT as if it were an authority on any important topic.

Perhaps GPT has passed the Turing Test, at least well enough that we are beginning to treat it as if it were an underperforming co-worker. It will be interesting to see what happens over the next few months.

What is ChatGPT and How Does It Work?

What is ChatGPT?

ChatGPT has been in the news a lot recently, and we’re going to hear a lot more about it in the near future. GPT stands for Generative Pre-trained Transformer, meaning that it uses a transformer network (a type of Deep Learning neural network) trained on a very large corpus of text to produce human-like text based on a prompt. GPT is a type of Large Language Model (LLM) and there are several other competing models in various states of development. One way to think of this is that GPT is a neural network implementation of a predictive text algorithm.

ChatGPT adds an interactive use interface to the front end of the GPT-3 model, allowing users to quickly and easily interact with the tool, and even allowing users to fine-tune how it responds based on previous interactions and examples of different writing styles. Give ChatGPT a prompt, and it will build an answer, an essay, or even a joke on a given topic by “analyzing” what has been written before in its training data and any data you have provided, and giving you a set of predictive text on the topic.

This post will break down what a predictive text algorithm is generally, what neural networks are and how they work, why ChatGPT is built using a neural network, and some background on ChatGPT’s development and how the underlying models were trained. The next posts will begin to look at some of the problems with LLMs, including GPT, and with ChatGPT specifically.

Predictive Text Algorithms

One way to conceptualize a predictive text algorithm is to imagine looking at every book that has ever been published (GPT also uses a much larger body of internet text, which presents additional problems that I’ll look at in a future post) and building a table of probabilities for every common English word (there are about 40,000 or them) that captures the odds that of what the next word will be. We get a table of 40,000 x 40,000 words with each cell filled with a probability representing how often in the training text the second word follows the first. As you might imagine, many of the entries will be zeroes.

With this table in hand, we can “prompt” the table by giving it a keyword, and then hop through the table by picking the entry with the highest probability of being the next word. If we ask our model to generate text based on the prompt “cat” we might get something like “cat through shipping variety is made the aid emergency can the.” Hardly a useful sentence.

In order to get an output that makes more sense (seems more like something a human would write), we might extend our table to include the probabilities for the most likely next word given the previous two words. Our output in this case would be a little better, but now our table has 64 Trillion (40,000^3 entries), rather than just 1.6 Billion (40,000^2). If we want to continue to improve the output of our algorithm, we need to continue it for greater depth of the-words-that-came-before. The problem with this approach is that it becomes too computationally expensive before it starts to produce good quality text. (For example, if we want to consider the previous 13 words, still not enough to produce a coherent essay-length output, we need a table with as one entry for each atom of hydrogen in 100 stars the size of the sun.) This problem is complicated by the fact that we also don’t have enough written text to fill out much of the table at that depth.

Because of these limitations, we need to find a better approach. We need an approach that is both more computationally efficient, and one that can achieve reasonable results using the training data we do have. GPT (and all Large Language Models, LLMs) uses a neural network to estimate the probabilities that we don’t have the computational capacity and necessary volume of training data to build traditionally.

Neural Networks

The idea of Neural Networks dates back to 1943, when Warren McCulloch and Walter Pitts formulated a simple model of networks of neurons and attempted to analyze the model mathematically. The first discussion of using neural networks as a possible approach to artificial general intelligence (AGI, or teaching machines to think “like” humans) was in the 1948 paper Intelligent Machinery by Alan Turing.  In 1949, Donald Hebb argued that such networks would be capable of learning. Computer simulations of neural networks began in the 1950s, but it soon became apparent that we didn’t have enough computing power to model enough neurons at a fast enough speed to do anything interesting. Neural Networks were put on the back burner of computing and AI for a few decades. They were studied rigorously again in the 1980s and 90s, but researchers decided that the only interesting things they could do were things that could be done more simply with probabilistic models. Neural networks finally broke out of the background between 2009 and 2012 when Swiss researchers successfully applied recurrent neural networks and deep feed-forward neural networks to win a series of international pattern recognition and machine learning competitions.

Neural networks work by simulating the type of computation that biological neurons perform. You can think of a simulated neuron as having a set of inputs (analogous to the connections (dendrites) coming to a biological neuron from other neurons); a function that converts those inputs to an output value (analogous to the body of the neuron); and a set of output connections that carries that output to some number of other neurons. Each output connection we can think of as a signal with a particular strength, or weight. In a neural network, these units are usually organized into layers, which are frequently (but not always) two-dimensional arrays of interconnected neurons. A number of layers are connected in ways that generate the type of output we are seeking. There are many different types of connection topographies within layers, many different types of layer architectures, and many different ways of “stacking” the layers. ChatGPT, which is built on GPT 3, uses something like 10 million neurons with about 175 billion connections arranged in about 400 layers.

Training the neural network is done by feeding it lots of data, and providing it feedback in the form of a difference signal from the desired output (at the final output layer, or sometimes at a layer close to the final layer). The network then adjusts the weights across the entire network in order to try to reduce the distance between the actual output and the desired output. As it turns out, it takes about as many training data runs as there are connections in the network to “fully” train the network. In the case of GPT-3, that means about 175 billion words of training data.

GPT-3 was trained on about 500 billion words of text. (Technically, tokens, which differ from words a little bit, but not enough to matter for our purposes.) The sources of that data are described in the table below (from Wikipedia):

Source: Wikipedia

Common Crawl is a dataset produced by the Common Crawl non-profit organization that contains over 3 billion we pages (as of Oct 2022). WebText2 is a corpus containing all Reddit submissions from 2005 through April 2022. Books1 and Books2 contain digitized published books. You may notice that the data sets are weighted withing the model in a way that isn’t the same as their token count. *Note: for a more detailed account of exactly how ChatGPT works, and an excellent overview of the underlying technologies, see this post by Stephen Wolfram: What Is ChatGPT Doing … and Why Does It Work?—Stephen Wolfram Writings

One of the things that neural networks are good at is generalizing patterns from very large, but incomplete data sets, more quickly and sometimes more usefully than a brute-force algorithm can accomplish. And that’s what ChatGPT does – it is a neural network that generalizes patterns from a large corpus of text and iteratively predicts what word should come next given the previous words (including the text it generated). By doing so, it has gotten to a level of capability that looks like useful human-written text in many ways, though it still has some important and significant limitations.

In the next post, I’ll look at some of the problems with GPTs and LLMs generally, and some of the specific issues with OpenAI’s approach to ChatGPT (and the underlying GPT-3 and -4 models).

Update:

As I was editing this post, OpenAI has released GPT-4 to the public. OpenAI has chosen not to reveal the specifics of the neural network implementation of GPT-4, but has promised that it will be “safer and more useful” than previous versions. Rumors, which have been denied by OpenAI’s founder, suggest that the model is on the order of 1000x bigger than GPT-3.

A Tentative Theory about Why the New Testament is so Revolutionary

As many of my recent blog posts do, this one starts with: I was recently having a conversation with a brilliant friend, and I was excited when I was forced to think about something in a new way. I love that feeling! Not only do I get to play with a new idea, I get to do so while sharing it with a treasured friend!

The conversation had made its way to our religious upbringings, although I don’t remember exactly how we got there; part of the joy of discovery. My friend Lyn was raised in the Catholic Church; I grew up mostly in the Episcopal Church of the late 70s and early 80s, and went to a couple of years of Catholic school, so we had much common ground. Lyn made an observation that immediately resonated with me: “I never quite understood how Christ’s suffering was supposed to absolve me of my sins, 2000 years later.”

It was an Aha! moment – I have frequently felt the same way, though I hadn’t ever articulated it so clearly. And as we shared our experiences with the popular interpretations of this idea, the idea itself seemed to get weaker and weaker. So many people seem to take the idea of Christ’s sacrifice as a license to live life how they want to, secure in the knowledge that their sins will be forgiven, so no problem! Isn’t that convenient?

As a “not an actual historian” who likes to read history (see the Story of Civilization series of posts, soon to be continued) and as a “not as actual Bible scholar who hasn’t read the whole book yet” who is interested in how religion helps weave the tapestry of culture, I took a stab at an idea about what the real value of the Christ story is, and what makes it so historically unique, and indeed revolutionary, that it remains powerful even in diluted and misunderstood forms two millennia later.

A Brief, and Probably Inaccurate, History of Polytheism

For many millennia, before the advent of writing and of history itself, mankind told stories about the gods. A frequent modern interpretation goes something like this: “people in the old times weren’t very smart, so when something they didn’t understand happened, or something was scary, they made up silly little kid stories about unseen and very powerful gods to explain it.” Of course, I think our ancestors deserve quite a bit more credit than this rather condescending view affords them. The same Greeks who had a multitude of gods and myths knew the Earth was round and measured its circumference to within 1% of the actual figure. With a stick. (Eratosthenes) The same Egyptians who had a multitude of (very similar) gods and myths built architecture more beautiful than anything built today, and aligned the pyramids to geographic north with incredible precision, again within 1% (Imhotep and others). Never mind that modern people are so mystified with how the pyramids were built that many people rely on a deus ex machina (Aliens!) to explain it. Similar examples can be found in every major culture, whether in India, China, Persia, the Norse lands, etc.

So what would it look like if our ancestors were actually much wiser than we commonly give them credit for? We might interpret the old gods and myths something like this: complex stories, rich in symbolism, much of which few modern non-scholars recognize, that served to transmit the wisdom of the centuries about how to navigate the difficulty and complexity of life. From this perspective, we see the purpose in each character, and the lesson in each story becomes richer and more meaningful. Yet we are left with gods, who although immortal and very powerful, are imbued at the deepest level with very human flaws and foibles. Zeus the philanderer. Hera the jealous and vengeful wife. Bacchus the immoderate partier who never grew up. Diana the goddess of both love and war (what a rich topic that is!). Vulcan, the forger of weapons with the tender heart whose gift is to appreciate beauty and whose curse is to remain outside its grasp. Hades, the guardian of the darkness. The cast of characters is nearly universal across cultures and across time. Yet in one respect, they aren’t exactly Gods as we think of them. They aren’t perfect beings distinctly different in character from humans. They aren’t superhuman; they are only super-humans. Human in temperament and character, but super powerful.

The Old Testament God

In my limited understanding of the Old Testament, we see a transition from polytheism to one God. But to me, the God of the Old Testament is again a super-human. He is jealous. He is vengeful. He demands sacrifice. He requires praise. Like the Greek gods, he frequently walks among humans on Earth. He judges whether the sacrifice is worthy and punishes swiftly and harshly if it isn’t. He wipes out entire cities. He turns people to pillars of salt. He even becomes so disgusted (is there a more human emotion than disgust?) that he wipes out all of humanity! Have we not, in the Old Testament, merely merged all of the old gods into one new, more powerful but still very human, god?

Of course, this is not to say that the Old Testament doesn’t contain nearly endless wisdom about the experience of being human. In fact, in its symbolism and story, it is extraordinarily rich and complex, and undeniably wise. Yet it seems that a large majority of the stories are meant to warn us of the perils of certain choices. The distance between an offering made from hard work being found unworthy to resentment, anger, vengeance, and nihilism is but 5 short verses in the tale of Cane and Able. How applicable to life, even today, that one should take notice when one finds himself feeling resentful, because that road can become very dark, very quickly, with great and harmful consequences. Likewise, the tale of Job reminds us again that sometimes virtue, instead of being rewarded, is punished, not just by other humans (as is so frequently the case), but by god and satan, or perhaps just the stark reality of life.

The New Testament Christ and God

Which brings us to the Christ story, which I recently heard described by Jordan Peterson as not only a tragic story, but technically speaking, the most tragic story possible. Here we have the most virtuous person, probably ever, who is being led to execution precisely because of his virtue. And he’s going to be executed in the most excruciatingly painful way possible, with a death deliberately designed by world’s global superpower to punish the worst of the worst. On top of that, he was betrayed by his friend. For money. Even worse, the authority in charge of the ordering his execution knows he’s innocent. And even worse, so does literally everybody else. And even worse, he is publicly shamed, mocked, and humiliated on the way to his execution. And even worse, Pilate offers clemency to one of the criminals being crucified that day. And that guy is guilty. And literally everyone knows it! Perhaps the only thing that I can think of that is more tragic than Christ’s experience is that of his mother, Mary, watching and witnessing all of this with grace. This act, of giving his life for humanity, is what gives rise to the interpretation that his sacrifice, specifically being punished precisely for his virtue, acts to cleanse humanity of our sins. But what of the rest of the Christ story? Might it fill the gap and also transform the Old Testament god into a different and new and more important and more useful character altogether?

In Christ, we find not just sacrifice, but we find an aspirational example, counter to the old gods in their flawed multitudes and the old god with his many, very human flaws. Through the journey of the Gospels, we witness an example, not of the perils of failing to properly balance the order and chaos in the world, but of how to live a virtuous life, to our full human potential in a way that improves not only our own lives, but our families’, our communities, and the world around us. In the process, the Old Testament god himself is transformed, and becomes a new God who is not a super-human, but who is superhuman in the sense of having transcended the inherent and unavoidable flaws of being human. And how revolutionary is that, that the Christ story transformed the nature of God himself?

We now have God as a loving father, a God who guides rather than demands, a God who forgives rather than enacting vengeance, a God of infinite patience who encourages us to thrive, a God who asks only one thing of us: do your best. And then gives us an example of what that can look like. And in so doing offers the challenge that it is hard work. And shows us the reality that we will be unfairly and perhaps tragically persecuted for it, not by him, but by life, by the condition of being human. And that we can handle that tragedy with grace. And that accepting the personal, individual responsibility for doing our best is worth it, precisely because it matters deeply, in the sense that each person who does so makes the world a richer place for everyone else. And in that way, God provides a purpose to each of us. A purpose not to please God, but a purpose that is pleasing to both God and to humanity.

And isn’t that way better than a life spent merely trying, unsuccessfully, to avoid tragedy and perhaps enjoy some momentary hedonistic happiness?

Hierarchy of Fitness Priorities

If you have a clear view of the purpose behind your fitness journey, and make actionable goals that align with that purpose, this hierarchy of priorities may be useful as you do the work, with discipline and intensity, necessary to improve your fitness, your lifespan and your healthspan.

In another recent conversation, this time on the social medias, I offered a framework for how I think about fitness training, which for me these days is mostly weightlifting, but also includes much walking and some other cardio and mobility work. I thought it might be useful to develop those ideas a little bit more and share my goals and my approach with this audience.

My Current Fitness “Goals”

As I recently shared in the What About Happiness post, I’m not the biggest fan of goals, per se, but it works well enough as a heading.

Early last year, February 28, to be precise, I had one of those moments in which I discovered that my ongoing rationalizations were bullshit. Specifically, I had been focusing on a specific strength goal, at the expense of overall fitness for about a year. And I’d gotten FAT. Not like Chris Farley fat, but more like John Belushi fat. Of course, I had rationalized that a little fat was to be expected when chasing a strength goal. And also, one can eat one’s way into a bigger bench press – weight gain translates very well to strength on that particular lift for most people. On that day I was up in the DC area for a dinner with some friends. The afternoon before the dinner, I went to the gym with one of my friends for a deadlift workout. I pulled a solid single, given my lack of focus on that lift at the time. But my friend Jack called me out – “nice pull, fat boy.” And he was right. The pull was ok, but damn, I was way fatter than I should have been.

The next day, I stepped on the scale. 252. I’m 5 ft 10 and I’m reasonably muscular from many years of lifting. But this was not good. And I had scheduled an anniversary cruise for July 31 with my wife. Being fat for cruise would definitely not be good. It was time to reframe and get serious.

What started as “I’m going to get as close to having abs as possible before cruise” quickly evolved into “I’m turning 52 this year. I’m fat. I have one chance to get in as good of physical shape as possible and then hang on for all I’m worth.” Later in the fall, witnessing some people close to me struggle with mobility issues and an injury from a fall, the importance of being fit and resilient came into very clear focus. In my teens and 20s, I think I overvalued physical fitness and undervalued mental and moral fitness. During my 30s and 40s, I think I mostly undervalued physical fitness, but made great progress on the moral and mental. It has now become clear to me that physical fitness is now at least as important to the pursuit of Eudaemonia as the moral and mental. The circle of life has come around, carrying with it added experience and a more mature perspective.

I started March 1, managing my own nutrition and exercise plans. Then with about 10 weeks to go before cruise, I got nutrition help from an IFBB Pro Figure competitor at our gym. By the time we left for our cruise, I had lost 30 pounds of fat. I’d probably also put on about 7-8 pounds of muscle. Didn’t quite get to a 6-pack. But we’ll call getting from a keg to a 2-pack progress. And I felt bold and confident on cruise, at the spa, and on South Beach. It was good to feel like the “in shape guy” again.

On the eve of the cruise…

After cruise, having just turned 52, I decided to test my deadlift again. I just wanted to see if I could still pull 10x my age. And it went up smoothly. So I added specific deadlift programming to my routine. If I could pull within 30 pounds of my lifetime best, 6 months after my last deadlift, well, then maybe it’s worth actually training the lift to see where I can take it. You know, one last time.

From January to July, I had been dealing with some shoulder issues in the shoulder I’d had repaired in 2016. Between the shoulder pain and the weight loss, my bench numbers collapsed. I finally saw an ortho doc in July, who helped me manage the pain so I could sleep through the night again, just in time for vacation. But his initial diagnosis of a rotator cuff tear wasn’t encouraging. Luckily, since then, I saw another ortho doc – and though there is a very minor tear, surgery isn’t required, and with a return to my rehab exercises, the shoulder feels better, and my strength is back as well. So, I added a bench press goal to my plan and started programming for that as well. You know, one last time.

So here we are on the back side of the holidays, and there was a bit of food binging and a few pounds of fat regained. Not a big deal, since it will be gone in a couple of weeks. But it’s time to recenter on goals again.

Long Term Goal: Get in as good of shape as I can this year, within the demands of the rest of my life, and build the discipline and habits to maintain/improve it for as long as possible. How I define this is roughly:

  • Sub 12% bodyfat (visible abs), to me, this is a marker of strong metabolic health and nutritional discipline, both of which are preventive for metabolic disease later in life (the deadly combo of heart disease, stroke, diabetes, Alzheimer’s)
  • Maintain strength within 10% of lifetime best and set new lifetime PRs on specific lifts. Maintain high lean muscle mass. For me, these are both important for maintaining the ability to participate fully in daily life as we age, and protective against falls and injuries in later life.
  • Get resting heart rate back under 60. Strong cardiovascular health supports avoiding metabolic disorders and also supports maintaining an active lifestyle
  • Return to running /jogging. I haven’t done these in many years, for a combination of reasons. But I used to enjoy them, and I don’t think I can be a fully capable dad and husband if I can’t run after dogs, grandkids (someday), or bad guys

Short-term Goals. Train specifically to hit lifetime strength goals in the bench press and the deadlift. Potentially compete in a master’s powerlifting meet in May. This will depend on a few factors. Not really important, but a nice to have.

Scott’s Hierarchy of Fitness Priorities. This is the real meat of the post I set out to write, despite the long pre-amble. I am frequently asked by both young gym-goers and peers how I approach training. The younger folks are usually looking for the ‘secret’; the more seasoned folks know the ‘secret’ doesn’t exist – so they ask for the ingredients. As fitness social media is full of top 3/5/10 things to do to achieve X type posts, I thought I’d take a little more general view of fitness and share what I think are the most important things to get right.

Here are what I view as the five most important ways to think about building and maintaining fitness:

1 (most important): Consistency. Don’t miss days. Even if you have to modify/scale back, don’t miss days. This isn’t to say that sometimes, due to life, illness, or injury, that wisdom won’t dictate that a day off is in order. But it does mean that discipline over long periods of time will win the game. As an example, Michael Phelps, after he became an Olympic champion, had a five-year period where he was in the pool at least once every single day. Every. Single. Day. The day after the Olympics? In the pool. Christmas day? In the pool. Every single day. This requires a certain mindset that seems to elude many people these days – discipline. One doesn’t have to want to do the work. One doesn’t have to feel motivated to go do today’s workout. One simply has to have the discipline to do what is necessary. It’s what adults do – we delay gratification because we know that the end result is better in the long run, for us and for those we care about. Be consistent.

Boatloads of progress are lost by people switching up their programs every month or two and switching up their nutrition four times a year or more (bulk/cut cycles). Once you have a program that aligns reasonably well with your goals (see below), stick with that program exactly as written for a year. After a year, you might be smart enough to make one minor change. A big part of consistency is allowing the process to work. Once the beginner gains are done, progress won’t be linear. Plateaus will happen. And you will delay or stall your progress more by changing a bunch of stuff all at once than by staying consistent and boring, doing the damn sets and reps, and letting the process do the work of breaking that plateau for you. When you become advanced enough to need advanced/elite training methods, you will know it. And it’s a lot later in the training journey than most people think. When the young guys in the gym ask, “how did you get that strong at bench press?” my answer is always the same. Five pounds at a time for many years. Be consistent. Effective training is almost always boring and repetitive.

2- Recovery. In my social media post, this was the nutrition section, but I think that was too narrow. Recovery includes sleep, nutrition, mobility, hot/cold treatment, body work, etc. Sleep – if your sleep is dysfunctional, it will severely limit your potential in every area of your life. It’s that important. Get a study and get it fixed. Nutrition. In our 20s, we could outrun the fork. In our 50s, we can’t. Good nutrition will improve progress more than anything else except fixing sleep. The beginners and intermediates in the gym talk about which program they do; the advanced guys and the big guys talk about nutrition. Mobility – have enough mobility to support good technique (see below) and prevent injury. If you don’t know where to start, go to a yoga class or google “Kelly Starrett Supple Leopard.” The others (hot/cold, body work) are mostly fine-tuning. If you discover that they help, and you have access to them, go for it!

3- Technique. Especially for people new to the weight room: let strength be a byproduct of practicing good technique. Because good technique will not happen whilst practicing strength. Be careful here. A lot of famous trainers are terrible and teach injurious techniques and exercises. Likewise, a lot of the instructional videos on the internet are terrible. See my How to Select a Personal Trainer post to increase your odds of getting good coaching. Perhaps I should consider a How to Find Good Fitness Advice on the Internet post, but I’m not sure I’m ready to go down that rabbit-hole. DM me if you want specific recommendations. There are some useful universal principles for good technique.

  • In lifting, like in boxing, all power starts from the floor. Build tightness from the floor up anytime you are lifting and your feet are touching the floor.
  • Likewise, learn how to brace your core – you can’t transmit force effectively through a shock absorber.
  • Learn how to grip the bar properly – this fixes a multitude of downstream problems, especially in the bench press.
  • Gravity acts 100% vertically – whenever possible stack the weight and your joints to apply force to the bar 100% vertically.
  • Know the difference between normal lifting movements and ballistic movements and keep these separate. Don’t try to make a curl a ballistic movement; don’t try to do slow-motion kettlebell swings.

Developing technique is a continuous process of attention to detail and learning. Roughly, it follows three phases. Beginners learn how to move the weight through a desired path of motion and develop the neuro-muscular coordination to generate efficient movement. Intermediate lifters learn how to apply the principles above to move their body through the desired range of motion and generate intensity to develop strength and muscle growth. Advanced lifters learn how to use the weight as a tool to put tension on the muscles in a specific way to generate a specific training stimulus in order to cause the muscle to adapt in the desired way. For a great example of this advanced intentionality in lifting, follow Mike O’Hearn on Instagram.

Make every single rep as technically masterful as you can. Use the best technique you can achieve on every rep of every set, especially warm-up sets. You can do one thing under light loads than you can’t do effectively under loads over about 85% of max – think. When the weight gets heavy, and you can’t think about more than one technique cue, your body will do what you taught it to do during all of your other sets. If you taught it to be sloppy during warm-ups, you will get sloppy under heavy loads. When you get sloppy under heavy loads, you will get injured. When you get injured, you blow up your consistency. When you blow up your consistency, you lose months or years of progress. Make every single rep, especially your warm-ups, as technically masterful as possible.

4- Intensity. Be intentional in your training. This can be hard sometimes, because effective training is almost always boring and repetitive. Learn how to generate brief periods of great intensity when the program calls for it. And learn how to turn it back off between sets. For cardio, learn how to focus on technique while hitting your HR zone with precision. Make it a game.

5- Programming. This is the sets and reps, the miles/yards/HR/intensity/intervals. Unless you have very specific goals in a particular domain (like a national level or higher competition), it matters way less than most people think. Find or build a program that aligns reasonably well with your goals and use it as the framework for doing 1-4. There are thousands of good enough programs for free on the internet and probably also thousands of great ones. Don’t major in the minors here. And don’t switch programs or nutrition plans until it is really necessary. See consistency above.

Conclusion. If you have a clear view of the purpose behind your fitness journey, and make actionable goals that align with that purpose, this hierarchy of priorities may be useful as you do the work, with discipline and intensity, necessary to improve your fitness, your lifespan and your healthspan.

In strength – S

How to Select a Personal Trainer

You are highly competent in most aspects of your life, but you want to get better at being physically fit. The problem is that 92% of personal trainers are terrible and it’s really hard to find one who isn’t. Use this guide to find a great fit.

Tips for success from a former personal trainer.

Tis the season for renewing commitments to personal growth. And for many people, that means a renewed commitment to “getting in shape,” whatever that may mean for each of us.

Yesterday, my friend Alex wrote a very thoughtful Facebook post about fitness that I engaged with, and that sparked a personal phone call based on some specific questions he had about his personal fitness journey. Because we live about 600 miles apart, and I can’t personally assess his fitness relative to his goals, I couldn’t offer any specific advice on exercise selection and progression. So, the conversation became less about his personal fitness goals, and more about how to select a personal trainer who can help guide him on his journey.

Who this article is for: In my time as a personal trainer, and in my decades of experience in competitive and recreational fitness, I’ve found that the people I see in the gym in my age group (born late 60s- early 70s), break down into roughly four groups:

  • The first group I call the Lost Boys/Girls. They go to a personal trainer, or more frequently a group fitness class, because either they want to do something for their fitness (something usually being better than nothing), or they want to be able to tell the story that they have a personal trainer/fitness group. Not that there’s anything wrong with that. I see these folks week in and week out, for many years, generally doing the same stuff at the same intensity and making little or no progress.
  • The second group are the Socializers. They seem to have a personal trainer because it gets them out of the house and they have someone to talk to for 30 minutes 3x per week. Not that there’s anything wrong with that.
  • The third group are the Gray-Haired Warriors. These are usually former competitive athletes of some variety who have a solid knowledge of how to train and eat for results. They set goals and make progress. They never work with a trainer, and almost never have a work-out partner. These lone eagles come in the gym, nearly every day, and just get the work done. They are typically in the top 5-10% in fitness in their gym, and for their age group, they are even more exceptional. They also have been in your gym long enough to know who the good trainers are and who the idiots are. Their advice is probably a cheat code to finding the right one.
  • The fourth group are the folks who really want to make progress toward some specific fitness goals, usually something like “I want to improve my health and fitness so that I can remain healthy and engaged in life as I get older.” But they don’t have the background of our gray-haired warriors and aren’t sure where to start. If they don’t find a Gray-Haired Warrior mentor or a really good trainer, they usually end up in the first group, but longing for more.

This article is for the people in the last group. You are highly competent in most aspects of your life, but you want to get better at being physically fit. The problem is that 92% of personal trainers are terrible, it’s really hard to find one who isn’t, and you want to avoid slipping into the purgatory of group one.

How to Select a Personal Trainer

So how do you find a great personal trainer? One who won’t just walk you to a series of machines and tell you to do 3 sets of 10, while they guffaw at the latest TikTok trending video and mumble the occasional empty word of detached praise? One who won’t try to indoctrinate you into their particular cult-like fitness niche, whether it matches your goals or not? One who won’t program idiotic, injurious movements with little benefit other than having been featured on the latest color glossy magazine cover?

Interview. I know those three free training session coupons are burning a hole in your pocket, but before your first training session, interview the trainer. After all, this is the person you want to help you achieve goals that are important to you, and not hurt you in the process. They work for you. A good trainer will be more than happy to discuss what your fitness goals are, why they are important to you, what your fitness history is, and what limiting factors you may have (injuries, old and new, mobility challenges, medical limitations, etc.). If they aren’t happy to have this conversation, move on to the next candidate. The trainer should listen actively, and should ask clarifying questions to ensure they understand. They should be able to restate what you told them. These sound like simple communication things (and they are), but they are critically important.

Red flags at this step: not willing to have the conversation, including we’ll talk about this after we get started; failure to listen actively and attentively; attempts to shoe-horn your goals into their preferred program.

Yellow flags at this step: offers simplistic advice without assessing your fitness (see next step); talks excessively about their credentials (beyond something like – yes, I can help you with that).

Assessment. After your prospective trainer passes your interview, schedule the first training session. A good trainer will use this first session to assess your ability across the relevant aspects of fitness relative to your goals. For weight lifting, they should assess your basic movement patterns, including at a minimum a basic squat movement, a basic hip hinge movement, a horizontal pressing movement, a vertical pressing movement, a vertical pulling movement, and a horizontal pulling movement. These assessments should be done initially with no load or minimal load (unless you have a decent training history – like a couple of years in the gym). A thorough assessment may require adding load to see where form begins to break, but should not approach a maximal effort. For monostructural cardio (running, biking, rowing, etc.), a few minutes of moderate Zone 2 effort should be sufficient to assess a beginning trainee. The goal of this training session is not to induce a training response. It is to assess the starting point.

Red flags at this step: trainer doesn’t do an assessment (this can be sorted out during the interview while discussing the first session); trainer tries to use machines to do the assessment (cable type machines are OK for pulling movements); trainer ignores any injuries you reported and insists that you do painful movements.

Yellow flags during assessment: trainer attempts to correct form during the assessment, beyond the minimal necessary instruction for a novice to perform the assessed movement.

Programming. At this point, the trainer should be ready to recommend a program for you to follow that takes into account your goals, your current abilities, and any limitations. The program should incorporate elements to help you progress toward your goals. I’m being pretty non-specific here, because there are a multitude of possibilities, depending on the person. Some principles to consider:

  • Is the program relatively simple? For most beginner and intermediate trainees, the program should be based on progressive overload and linear progressions. Unless you are an advanced or elite trainee, you don’t need wave periodization or more complex training methods. If you are advanced or elite, you know it already, and this article isn’t for you.
  • For weight training, since this is where most people have questions, the program should be primarily based on compound (multi-joint) lifts or appropriate variants for the trainee. These are the same types of lifts discussed in the assessment section – squats, hip hinges, horizontal and vertical presses, horizontal and vertical pulls. If you have limitations that prevent you from performing any of these with non-injurious form, then the trainer should be designing a routine with progressions of the main lifts and/or related exercises to build the primary lift. For example, if a trainee can’t get into a safe initial position for a deadlift because they are unable to maintain a neutral spine position, the trainer should be looking at something like 45 degree back extensions and core bracing exercises to build the ability to maintain a neutral spine position before introducing the deadlift.
  • Does the program integrate with your lifestyle? Is it achievable? For most people seeking general health and general purpose fitness, something like 2-3 days of weight training for 30-60 minutes, 2 hours per week of Zone 2 cardio, mobility work appropriate to the trainee, and maybe one session of high-intensity cardio (like a Tabata on the Air Bike) every week or two is more than sufficient.
  • Did the trainer discuss nutrition as part of the plan? Did they go deeper than calories-in/calories out? Did they discuss that scale weight is only part of how you look and feel, and that losing 5 pounds of fat and also gaining 5 pounds of muscle will make most people look MUCH better than just losing 5 pounds of fat? Did they discuss realistic expectations for long-term fat loss (about 1 pound per week is sustainable) and muscle gain (about one pound per month for non-enhanced trainees)?

More Red Flags. I mention these because I see them ALL THE TIME in commercial gyms, largely because magazines and influencers, who are chasing clicks and follows, have discovered that flashy, complicated, fancy stuff sells over boring and effective hard work.

  • Any trainer who wants you to deliberately be unstable with an external load (holding a weight of any type) should be summarily dismissed as your trainer. I could write an entire post about this particular flavor of personal trainer idiocy, but I won’t. I’ll just sum up thusly: 99% of the time they do this to improve the trainee’s balance, which they sometimes also sell as “strengthening the little stabilizer muscles.” While balance is an important component of fitness, I have found that being stronger naturally improves my balance, and if I need to work on balance as a skill there are programs, usually based in yoga or gymnastics, that are very effective without the disproportionate risk of doing balance under load. Why would you want to hold a heavy iron weight over some part of your anatomy, especially your teeth and throat, and then deliberately put yourself into an unstable position? Here are a few pictures of some of the dumb exercises I’m talking about:

If you are morbidly curious, you can find plenty of videos of people breaking ankles (bosu squats), breaking wrists when the ball pops (presses on the ball), and dropping the bar on their face/neck (feet up bench). In general, for beginning and intermediate trainees, these exercises are nothing but risk incurred for a reward that can be found elsewhere.

  • Trainer doesn’t understand physiology/kinesiology. The simplest way to spot this red flag is if the trainer is trying to get you to move the weight in an ideal path relative to the exercise itself, and not relative to your particular proportions and leverages. If you aren’t sure, and you are struggling to execute an exercise as the trainer is instructing, ask the trainer: is there something I should be doing to adjust the exercise due to my proportions? They should be able to explain how your build/proportions/physiology relate to the exercise and what adjustments, if any, are necessary. If they pass this, then it may be that you don’t yet have the training experience, proprioception, or body awareness to translate their instructions into the actions they are seeking, and the trainer should adjust their cuing to help you make progress. If they get frustrated and head down the path of “you just don’t get it” or “you aren’t very coachable” or “I’ve never seen someone struggle with this before,” time to find a new trainer.

Conclusion. Selecting a trainer is difficult, because most of them are unqualified for the job, and those that do have official certifications are frequently still not skilled enough to help you without creating unnecessary risk. So the burden falls to you to find a good trainer. I’d encourage anyone to ask the advice of the most fit people in your gym – they will have a pretty good idea, and probably some pretty strong opinions, about who to avoid and who to seek out. Then apply this template to ensure a great fit.

In strength- S

What About Happiness?

Goals, Systems, and Identity

I was privileged to have a great conversation with a brilliant friend recently. Emily is someone who is an exceptionally high performer. She’s at the top of her chosen profession. She has a successful and fulfilling marriage. She also runs a non-profit foundation. She has two kids who are doing very well. She runs a global mentoring network. She exercises, meditates, and prays daily. She also is on the school board, church board, neighborhood board… I probably left out a few things.

Her observation was that she has achieved every goal she has ever set for herself. She’s wise enough to understand that the universe gets a vote in those goals; that is to say that the outcomes weren’t fully in her control and that to some degree luck and fortune amplified her efforts. Yet, she finds herself struggling with a really hard question: I’ve achieved every goal I’ve ever set for myself – now what? I don’t want to keep chasing goals to find happiness.

A voracious reader, she set out to look for answers and read several current books on the topic. But still, something wasn’t connecting.

In one of those moments where an insightful question causes the listener to connect dots previously left unconnected, we stumbled on an idea. Perhaps there’s a progression to this, and Emily’s expectations haven’t quite caught up with her progress.

We started with goals -discrete problems we will solve on a particular timeframe, such as “I will lose 10 pounds by Memorial Day.” The problem with goals is that happiness lives on the other side of achievement. So we set a goal, we paint a picture of how happy we will feel, and how we will feel happy, when we attain our goal. This picture serves as motivation to get started, and provides the ‘why’ to generate the dedication to stick to it.

Everything is going great until… the day we achieve the goal. Perhaps we give ourselves a moment of the happiness we imagined. But very quickly, one of two things happen. Having arrived, we set a new goal. Or worse, we celebrate by reversing our progress and have to set the same goal again! (For instance, celebrating losing weight by indulging in pizza and cookies!). Very quickly, happiness moves back over the horizon of goals. Of course, it is also possible that we never achieve the goal – either due to factors within our control or factors outside of our control. Feeling that we can’t achieve happiness because we can’t achieve our goals because of factors outside of our control can lead to some very dark, very nihilistic places. Not good.

Scott Adams famously (but probably not originally) suggested a solution. Implement systems rather than goals. James Clear made essentially the same recommendation in the very popular book Atomic Habits. Instead of “I will lose 10 pounds by Memorial Day,” a systems statement lives a level of abstraction higher. A systems statement might be of the form “Every day, I will exercise in some way, even if it is just a brisk walk.” The systems idea is that if one builds habits and patterns, then those habits create opportunities to succeed at a variety of pursuits. And some of these might not be opportunities that one could plan and make goals for, because the future is too unpredictable. A famous example is Steve Jobs. Jobs took classes in college that he thought were interesting, even though he didn’t know whether they would be useful in the future, and if so, how. He thought calligraphy was one of those interesting things that he wanted to learn more about. Because one of Jobs’ systems was to learn new things about diverse topics, he was able to connect his knowledge of calligraphy with his knowledge of computers and his brilliance at marketing to make the first PC with proportional type fonts – the MacIntosh. But Jobs couldn’t have planned to build the first Mac and then worked that plan into a list of goals, one of which would have been ‘learn calligraphy while in college.’

Where is the happiness in the systems model? I think it has to be in the joy of creation. The feeling of ‘Aha!’ when ideas and skills, previously separate, combine to create something new and, hopefully, valuable. So what’s the problem with that? Well, most people want to feel like they are working toward something meaningful. We are stimulated by challenges because we know the direction our efforts are leading, if not the end result. But if all we feel are challenges, without direction or the guarantee or even probability of that meaningful ‘Aha!’ it might be a lonely, barren journey. Even worse, it could devolve into the “Ants Marching” (Dave Matthews Band) problem wherein we do everything we are expected to do, with no hope of joy at the end of the journey. Another dark place.

Serendipitously, while we were discussing the systems approach, I mistakenly phrased an example as “I am the kind of person who exercises every day.” And here was a fun ‘Aha!’ moment. This is an identity statement, not a systems statement. It isn’t about a habit, it is about being or becoming a particular type of person. Emily quickly recounted how her mother used identity statements to help her succeed. If she was struggling in math, her mother would remind her that “you are the type of person who can do well at this.” And then when she succeeded “of course, you are a person who is good at math!” Brilliant!

Identity statements quickly led us to remember Aristotle’s virtue ethics, because Artistotle’s virtues can be phrased as the type of identity statements most people would like to aspire to. For example: “I am an honest person.” Or: “I am a courageous person.” Or even better: “I am a prudent person – I deploy the right virtues correctly in the appropriate situations.” Where lies happiness in the virtue model? The Greeks had thought about this so deeply that they had a special word for it: Eudaemonia. No such word exists in English. The closest might be “human fulfillment.” Even better might be “the feeling of deep fulfillment that comes from deliberately living a virtuous life.”

So our tentatively-developed model seems almost like a set of conditions or ways of living one might be able to progress through. From goal-setting and achievement, with its brief and fleeting rewards, to building systems and habits for improving ourselves and helping others, with the attendant greater but unpredictable joys, to practicing an identity as someone who consciously pursues becoming the most virtuous and capable version of ourselves, with the result of achieving human fulfillment. In this model, Emily was consciously in the systems mindset, although she rediscovered that she already knew the identity/virtue mindset, and she realized that she was seeking the wrong feedback system. Emily discovered that she was still trying to measure happiness, even though she is already well along the way to fulfillment. To Eudaemonia.