The Long Game 166: Parenting Time, Protein Powders, Discomfort, LLMs, Consciousness
đ§Ș Lead in Your Shakes, Stoic Discomfort, AI Brain Rot, Quantum Memory, Love as Dialogue, and More
Hi there, itâs Mehdi Yacoubi, co-founder at Mirage Metrics and Mirage Exploration, and this is The Long Game Newsletter. To receive it in your inbox, subscribe here:
In this episode, we explore:
Parenting time
Protein powders and lead
Discomfort is the price you pay
LLMs are a different kind of intelligence
Your consciousness can jump through time
Letâs dive in!
đ„ Health
đ§Ș Protein Powders and Shakes Contain High Levels of Lead
Roughly 6-10 years ago, we began to see a growing number of doctors recommending increased protein intake for health and longevity, which led to pushing the 1g/lb bodyweight recommendation (the classic bodybuilding recommendation) to the wider population.
Everyone who has tried or is eating 1g/lb of bodyweight of protein every day knows itâs not easy to hit this number.
For example, at 220-240 (depending on the period), I should be eating 220+g of protein per day, which is obviously a lot of protein.
This led many people to drink a good share of these proteins in a liquid format, with protein powders exploding in popularity.
Last week, this report showed that many popular brands of protein powders contain high levels of lead.
Much has changed since Consumer Reports first tested protein powders and shakes. Over the past 15 years, Americansâ obsession with protein has transformed what had been a niche product into the centerpiece of a multibillion-dollar wellness craze, driving booming supplement sales and spawning a new crop of protein-fortified foods that now saturate supermarket shelves and social media feeds.
Yet for all the industryâs growth and rebranding, one thing hasnât changed: Protein powders still carry troubling levels of toxic heavy metals, according to a new Consumer Reports investigation. Our latest tests of 23 protein powders and ready-to-drink shakes from popular brands found that heavy metal contamination has become even more common among protein products, raising concerns that the risks are growing right alongside the industry itself.
For more than two-thirds of the products we analyzed, a single serving contained more lead than CRâs food safety experts say is safe to consume in a dayâsome by more than 10 times.
âItâs concerning that these results are even worse than the last time we tested,â said Tunde Akinleye, the CR food safety researcher who led the testing project. This time, in addition to the average level of lead being higher than what we found 15 years ago, there were also fewer products with undetectable amounts of it. The outliers also packed a heavier punch. Naked Nutritionâs Vegan Mass Gainer powder, the product with the highest lead levels, had nearly twice as much lead per serving as the worst product we analyzed in 2010.
Even in the lifting and bodybuilding niche that I follow a lot, you can start to feel the tide is shifting. Many people realize that eating more protein than they need takes away from useful carbs that really make the difference when you train.
I think .7 to .8 g/lb of bodyweight is likely to be a better recommendation, at least this is my current target.
Pair with: Why You Donât Need High Protein and Top-rated protein powder (if you insist on drinking protein shakes), and ArrĂȘtez de vous prendre la tĂȘte avec les protĂ©ines ! by my friend NFKB.
đ± Wellness
âł Parenting Time
I remember watching a news show right when my daughter was a few months old, and the host said something along the lines of âif you complain about all the infinite requests of your baby battling for your attention, if this is annoying to you, then you will almost certainly not enjoy the presence of your children once they become adults.â
This graph was a brutal reminder of how quickly it passes.
More than 40% of your parenting time has elapsed by the time your kid enters kindergarten.
Once they get into middle school it is 2/3rds gone.
Pair with: Your life in weeks and Vitalism, to hopefully make this graph much, much longer.
đ§ Better Thinking
đ
Discomfort is the price you pay for a fulfilling life
I could not agree more with the core idea of this piece. It complements perfectly the ideas on the shortcomings of the over-optimization lifestyle Iâve been sharing here for some time.
I find myself asking this more and more: since when did boundaries and routine become more important than growth, adventure, and being there for others? Why do some people dread going to the plans they made, and would rather go to bed early so they can wake up for their morning routine? I wonât lie, Iâve been a victim of this. I think after the pandemic, I may have started âprotecting my peaceâ a bit too much. Thankfully, I snapped out of it. Grabbing coffee to catch up with a friend and not missing someoneâs celebration is more important than being able to wake up at 6 a.m. to get 10,000 steps in before 9 a.m.
Good ideas for all of us:
Get familiar with the unknown. The feeling of exploring a coffee shop youâve never been to, the initial awkwardness of talking to a stranger, the unease of eating alone at a restaurant, the effort of dragging yourself off the couch to go dancing with your friends, the indulgence of letting yourself sleep until 11. Thereâs more to life than constant stability. When your days start blending together, make it a point to do at least one thing to break up the monotony: walk a different way to work, put on that dress thatâs been hanging in your closet for a âspecial occasion,â or book the trip.
To ponder:
Itâs all connected: the relationships you nurture, the hobbies you pursue, your travels, your daily life. Youâll stumble, feel exhausted, and be hard on yourself. But your patience and persistence will ultimately pay off. Leaning into that discomfort and pushing yourself, whether itâs setting aside an hour to write, reworking a painting again and again, spending hours perfecting an outfit, or saying yes to last-minute plans, will ultimately feel more rewarding because you didnât give in to those limiting beliefs.
Pair with: My generation isnât suffering enough
There are ways to practice the best of both worlds. One is the ancient stoical practice of voluntary discomfort: the art of suffering on purpose. This involves integrating simple yet unpleasant tasks into your life, say taking a cold shower for 30 seconds each morning, adopting a strict exercise regime, or giving up alcohol. Extreme athlete Wim Hof has even designed a simple breathing technique to foster resilience, involving voluntary activation of the sympathetic nervous system, a method which has been linked to health benefits such as reducing stress, boosting the immune system, and easing the symptoms of autoimmune disorders. Others encourage intermittent fasting, a method which has also been linked to reduced anxiety and depression. Voluntary discomfort doesnât have to be physical, either. It may involve intellectual unease: reading things you disagree with, engaging in arduous debates, trying to dismantle and re-sculpt your worldview.
đ€ AI Updates
đŹ LLMs are a different kind of intelligence
Karpathyâs latest interview was very good. I highly recommend it.
This post complements it well:
The most interesting part for me is where @karpathy describes why LLMs arenât able to learn like humans.
As you would expect, he comes up with a wonderfully evocative phrase to describe RL: âsucking supervision bits through a straw.â
A single end reward gets broadcast across every token in a successful trajectory, upweighting even wrong or irrelevant turns that lead to the right answer.
> âHumans donât use reinforcement learning, as Iâve said before. I think they do something different. Reinforcement learning is a lot worse than the average person thinks. Reinforcement learning is terrible. It just so happens that everything that we had before is much worse.â
So what do humans do instead?
> âThe book Iâm reading is a set of prompts for me to do synthetic data generation. Itâs by manipulating that information that you actually gain that knowledge. We have no equivalent of that with LLMs; they donât really do that.â
> âIâd love to see during pretraining some kind of a stage where the model thinks through the material and tries to reconcile it with what it already knows. Thereâs no equivalent of any of this. This is all research.â
Why canât we just add this training to LLMs today?
> âThere are very subtle, hard to understand reasons why itâs not trivial. If I just give synthetic generation of the model thinking about a book, you look at it and youâre like, âThis looks great. Why canât I train on it?â You could try, but the model will actually get much worse if you continue trying.â
> âSay we have a chapter of a book and I ask an LLM to think about it. It will give you something that looks very reasonable. But if I ask it 10 times, youâll notice that all of them are the same.â
> âYouâre not getting the richness and the diversity and the entropy from these models as you would get from humans. How do you get synthetic data generation to work despite the collapse and while maintaining the entropy? It is a research problem.â
How do humans get around model collapse?
> âThese analogies are surprisingly good. Humans collapse during the course of their lives. Children havenât overfit yet. They will say stuff that will shock you. Because theyâre not yet collapsed. But we [adults] are collapsed. We end up revisiting the same thoughts, we end up saying more and more of the same stuff, the learning rates go down, the collapse continues to get worse, and then everything deteriorates.â
In fact, thereâs an interesting paper arguing that dreaming evolved to assist generalization, and resist overfitting to daily learning - look up The Overfitted Brain by @erikphoel.
I asked Karpathy: Isnât it interesting that humans learn best at a part of their lives (childhood) whose actual details they completely forget, adults still learn really well but have terrible memory about the particulars of the things they read or watch, and LLMs can memorize arbitrary details about text that no human could but are currently pretty bad at generalization?
> â[Fallible human memory] is a feature, not a bug, because it forces you to only learn the generalizable components. LLMs are distracted by all the memory that they have of the pre-trained documents. Thatâs why when I talk about the cognitive core, I actually want to remove the memory. Iâd love to have them have less memory so that they have to look things up and they only maintain the algorithms for thought, and the idea of an experiment, and all this cognitive glue for acting.â
These quotes are also a good overview of the episode:
1. LLMs donât work yet
They donât have enough intelligence, theyâre not multimodal enough, they canât use computers, and they donât remember what you tell them. Theyâre cognitively lacking. Itâll take about a decade to work through all of that.2. When you boot them up, they always start from zero
They have no distillation phase, no process like sleep where what happened gets analyzed and written back into the weights.3. Whatâs stored in their weights is only a hazy recollection of the internet
Itâs just a compressed blur of 15 trillion tokens squeezed into a few billion parameters. Their context window is just short-term working memory.4. Theyâre good at imitation, terrible at going off the data manifold
Too much memory, not enough reasoning. We need to strip away the memorized knowledge and keep the cognitive core: the algorithms, the magic of intelligence, problem-solving, strategy.5. Weâve probably recreated a cortical tissue, pattern-learning and general, but weâre still missing the rest of the brain
No hippocampus for memory. No amygdala for instincts. No emotions or motivations.6. They memorize perfectly but generalize poorly
If you give them random numbers, they can recite them back. No human can do that. Thatâs the problem: humans forget just enough to be forced to find patterns.7. Anything truly new, code thatâs never been written before, ideas that have no template; they stumble
Theyâre still autocomplete engines with perfect recall and no understanding. Until we find that cognitive core, intelligence stripped of memory but full of reasoning, theyâll stay brilliant mimics, not minds.
It also confirms what Iâm seeing on the ground every day: AI is great and has a lot of potential, but the impact and speed of change coming from it is overblown.
Pair with: LLMs can get brain rot
This is crazy, but just like us, LLMs can get brain rot:
We propose and test the LLM Brain Rot Hypothesis: continual exposure to junk web text induces lasting cognitive decline in large language models (LLMs). To causally isolate data quality, we run controlled experiments on real Twitter/X corpora, constructing junk and reversely controlled datasets via two orthogonal operationalizations: M1 (engagement degree) and M2 (semantic quality), with matched token scale and training operations across conditions.
Contrary to the control group, continual pre-training of 4 LLMs on the junk dataset causes non-trivial declines (Hedgesâ g>0.3) on reasoning, long-context understanding, safety, and inflating âdark traitsâ (e.g., psychopathy, narcissism). The gradual mixtures of junk and control datasets also yield dose-response cognition decay: for example, under M1, ARC-Challenge with Chain Of Thoughts drops 74.9 â 57.2 and RULER-CWE 84.4 â 52.3 as junk ratio rises from 0% to 100%.
Error forensics reveal several key insights:
Thought-skipping as the primary lesion: models increasingly truncate or skip reasoning chains, explaining most of the error growth.
Partial but incomplete healing: scaling instruction tuning and clean data pre-training improve the declined cognition yet cannot restore baseline capability, suggesting persistent representational drift rather than format mismatch.
Popularity as a better indicator: the popularity, a non-semantic metric, of a tweet is a better indicator of the Brain Rot effect than the length in M1.
Together, the results provide significant, multi-perspective evidence that data quality is a causal driver of LLM capability decay, reframing curation for continual pretraining as a training-time safety problem and motivating routine âcognitive health checksâ for deployed LLMs.
âĄïž Startup Stuff
đ§ââïž AI Didnât Kill Tech Jobs, it Killed the Ladder
This piece shares very good ideas for young people looking for AI & tech jobs.
First, the problem:
AI didnât destroy the tech job market; it just massively changed it. This is not just an opinion; itâs facts. Letâs talk numbers.
AI didnât kill tech jobs, it stumped on entry-level roles. A Stanford study shows a 13% decline in early career roles that were exposed to AI. On the other hand, the demand for more experienced senior roles saw a 6% increase. You can read this blog post by Sundeep Teki on the 6 major ways AI changed the software developer job market for more details.
Then the solution:
Start practicing today, so youâre not stressed tomorrow.
Build your portfolio (vibe-coding your app, contributing to open-source, anything you can think of), share what you learn on X, get high-quality connections, and enjoy the process.
The time for unpaid internships is gone, when you can have 100% control over what to work on. Entry-level roles are also on a decline. Youâre on your own to get yourself going, and nobody is coming for help.
There has literally never been a better time to work independently on yourself. AI gives you such leverage that people didnât have 5 years ago. A lot of smart guys out there are using AI to enhance what they can do, rather than compete with it. And in 5 years, those will make it out successfully, that have teamed up with AI to make cool stuff and ship it.
I see it from personal experience: Iâm much more likely to want to talk with a young engineer/ builder if they just send me a cool side project they did.
Pair with: Build your personal moat
My favorite career advice is to develop a âpersonal moat.â
A personal moat is a set of unique and accumulating competitive advantages in the context of your career.
Like company moats, your personal moat should be a competitive advantage specific to you thatâs not only durable, but compounds over time.
This should be something thatâs:
Hard to learn and hard to do (but perhaps easier for you)
Impossible without rare and/or valuable skills
Unique to your own talents & interests
Legible, in the sense that your expertise should be easy to describe, easy to share, and makes people want to do both for you
đ What I Read
đ The Great Feminization
Thought-provoking piece.
Women can sue their bosses for running a workplace that feels like a fraternity house, but men canât sue when their workplace feels like a Montessori kindergarten. Naturally employers err on the side of making the office softer. So if women are thriving more in the modern workplace, is that really because they are outcompeting men? Or is it because the rules have been changed to favor them?
Pair with: Rise of the Beta
đ€ Populism Fast and Slow
A very good piece on the issues with the elites.
Unfortunately, there are many cases in which the people are right to distrust elites. Analytical reasoning is sometimes a poor substitute for intuitive cognition. There is a vast literature detailing the hubris of modern rationalism. Elites are perfectly capable of succumbing to faddish theories (and as we have seen in recent years, they are susceptible to moral panics). But in such cases, it is not all that difficult to find other elites willing to take up the cause and oppose those intellectual fads. In specific domains, however, a very durable elite consensus has developed. This is strongest in areas where common sense is simply wrong, and so anyone who studies the evidence, or is willing to engage in analytical reasoning, winds up sharing the elite view. In these areas, the people find it practically impossible to find allies among the cognitive elite. This generates anger and resentment, which grows over time.
Pair with: A Psychological Theory of the Culture War
This article presents a psychological theory of the culture war, and posits a dynamic social system in which the actions, rhetoric, and behaviors of each side influence the other. People are not seeking their own economic interests nor even working towards a moral vision, but responding to a built-in drive towards trying to achieve status, which involves tearing others down. Itâs something of a LARP because those who are most unaware of their own motivations can act with the most certitude, and therefore have the largest effects on our political culture.
đ° The Making of a Market Maker
How Thomas Peterffy built the machines that killed the trading floor and made Interactive Brokers into a $100 billion business.
So many things to learn here. Some of the learnings are that he never read a business book (I actually think this is great advice) and that he still owns 75% of his business, showing how important keeping control over your business is (same as we see with Larry Ellison, Dyson, etc.) It might seem unimportant in todayâs world where people raise round after round, diluting themself to very low ownership with the idea of becoming a decacorn. But having a huge business with low control over it is not necessarily the position you want to be in.
I brought up Costco, a comparison investors like to make. Both companies are built on the radical notion that you can make more by charging less. One of Peterffyâs neatly groomed eyebrows lifted.
âIâve never been to Costco,â he said.
I asked if he had studied other businesses, or learned from mentors. The question seemed to exhaust him. His eyes closed briefly.
âIâm sorry,â he said. âCan you repeat that?â
When I did, he replied: âIâve never read a business book.â
In 2019, on his 75th birthday, he stepped down as CEO, handing the reins to Milan Galik, a Slovakian software engineer who joined the company in 1990. Peterffy gave up horse riding at 70 and skiing after a fall, but retirement is out of the question. Heâs 80, chairman of Interactive Brokers, still owns nearly 70% of the business, and says heâs âsort of running the sales and marketing department because nobody wants to do it.â
Pair with: John Collison in conversation with Stanley Druckenmiller, and how many great firms are maniacal about the information they consume.
đ The key to love is understanding
On relationships, conversations, listening and understanding:
endless conversation as a love story
If life goes well for me, I hope my dream relationship will just feel like one long, continuous conversation. I donât mean that metaphorically, but literally â a never-ending dialogue of becoming. I want to walk and talk, lie in bed and talk, sit across from each other at dinner and talk. I want to reflect on our days together, ask strange questions, spiral into metaphysics, laugh about dumb things, and learn the specific, rippled texture of each otherâs minds, studying their soulâs cartography, one contour at a time.
Voice and language so precise and personal that when one of us dies, as is the nature of life, the conversation doesnât end. It continues in the empty space, in the wondering: What would they have said? The dialogue stretches into the beyond, because conversation lives on and on. This is romance, to me.
To me, this is intimacy: mutual prompting, mutual witnessing, an endless back-and-forth of seeing and being seen, a daily work of co-created shared language. The people we love participate in our becoming, as much as they observe us. They reflect parts of us we didnât know were there. Their curiosity rearranges us and their listening sharpens us.
Pair with: Masters of Love
đ Brain Food
đ€Ż Your Consciousness Can Jump Through Time
This one is crazy:
On an early October night in 1989, a four-year-old girl was shocked awake by a phone call and a scream. She tiptoed barefoot on the clammy vinyl tile of the hallway. âHe died in a car accident!â her motherâs voice cracked before it shattered. The girlâs shining dark eyes could only stare. From the moment she threw her arms around her father before he boarded his flight for that fateful business trip, she knew she would never see him alive again.
This is just one of the myriad and often eerie accounts of precognition that have been shared with cognitive neuroscientist Julia Mossbridge, Ph.D. But it was her own experience with these strange, psychic âgut feelingsâ that led her to study them in the first place.
As far back as the age of seven, Mossbridge has had precognitive dreams, she says. She and her parents were skeptical of them until she began recording the details in a dream journal. While she admits sheâs misremembered some of her dream visions, sheâs also been able to foretell events from the future that she would have had no other way of knowing.
She says these memories from the future could mean the notion of time might not be as linear as we imagine.
And a possible explanation:
Precognition could be explained as a form of quantum entanglement, Radin says. Particles that are entangled are supposed to share the same information and behave the same way, even from far away, which is what Einstein called âspooky action at a distance.â Radin thinks this theory might explain why we can remember things that have not happened yet.
âSome people hypothesize that precognition is your brain entangled with itself in the future, because entanglement is not only things separated in space, but also separated in time,â he explains. âIf it can be entangled with itself in the future, in the present youâd be feeling something like a memory that is going to happen in the future.â
If time is not so linear and consciousness can enter an invisible portal to the future, it might explain the feeling of dĂ©ja vĂ». Regardless, the phenomenon of precognition is backed by statisticsâitâs just a matter of proving what the mechanism could be, Mossbridge says.
I am convinced that there is much more to reality than we currently understand, so I do not like to dismiss seemingly âcrazy-soundingâ ideas.
Pair with: CIA archives on precognition and Reality is an illusion
đ„ What Iâm Watching
đČđŠ Exploring the Souk in Tangier
I love Tangier, I was there a few weeks ago, and enjoyed this OG Bourdain episode.
Pair with: The movie Only Lovers Left Alive
đ§ How To Force Your Brain to Crave Doing Hard Things
This guy will literally rewire your brain to stop procrastinating.
đ§ The Tool of the Week
đ Reign Energy Drink
Iâm struggling to find good energy drinks in Europe. It seems they canât legally have the same caffeine content as the US ones. I was in France a few days ago, found a Celsius, bought it, only to discover later it has only 115 mg of caffeine đ
My favorite option is Reign, with 200mg, but itâs rare to find.
đȘ Quote Iâm Pondering
âAbove all, do not lose your desire to walk. Everyday, I walk myself into a state of well-being & walk away from every illness. I have walked myself into my best thoughts, and I know of no thought so burdensome that one cannot walk away from it. But by sitting still, & the more one sits still, the closer one comes to feeling ill. Thus if one just keeps on walking, everything will be all right.â
â SĂžren Kierkegaard
đ EndNote
Thanks for reading!
If you like The Long Game, please share it or forward this email to someone who might enjoy it. You can also âlikeâ this newsletter by clicking the â€ïž just below, which helps me get visibility on Substack.
Until next time,
Mehdi Yacoubi
Lots of great stuff to chew on this w/e