Dave Lee

Host of YouTube channel Dave Lee on Investing, TSLA investor since 2012, building with AI.

Shift or Noise - new series of 3-minute AI news videos with my take

[Dave Lee @davelee]

This year has been a trying year for my family. Wife has had to deal with surgery, chemo, and now radiation for stage 3 breast cancer.

To help my kids cope and thrive I’m having my kids hit a lot of balls everyday (table tennis and tennis). It’s part distraction but it’s also a way for them to grow in their coordination, reflexes, athleticism, and more. It keeps them positive, active and eager to learn and grow.

Focus on the ball and hit it. Hundreds of times everyday. Something good will come out of it. At least that’s my hope.

[Dave Lee @davelee]

Fable 5.2 was released yesterday. Question is how big of a jump is this new model? What can it do that previous models couldn't?

My intro:
"Because new AI models get released often, it's easy to get numb to it and just overlook what's going on. But I think it's important to pay attention, and here's why.
These AI models are generalized intelligence, meaning they're not just narrowly applied to math or English. Many of the capabilities of the newest, most powerful models aren't even fully known or discovered. In other words, you can apply it to solve a new problem that people didn't even know it could do. There's just too many possibilities for the AI companies to know them all."

My ending:
"AI is not just getting smarter, but it's also taken a huge leap in how it can work together with other AI.
Basically, AI can work together as a large thousand-AI-agent company to work through one of the hardest proofs in math, and get every step checked by a computer.
But that's just the beginning, since we're in the infancy stages of AI. AI will just get smarter and more capable. And it can tackle increasingly more difficult problems.
What if a million AI agents tackle cancer or Alzheimer's? And what if those million AI agents get a huge intelligence upgrade every two months? How long will it take to find breakthrough discoveries and cures?
But it's also interesting because there's a few questions. Who has the compute power and resources to run a million AI agents and direct them to work they want to do? Which big problems are the most likely next candidates?
And another question is: is it even practical for a non-frontier-lab company to run a million AI agents of the latest and greatest AI model, when the frontier lab companies can run them up to ninety percent cheaper than what they charge others?
In other words, the frontier lab companies can run them at cost, while others have to pay a large markup, which can make things prohibitive.
And then there's the question of the frontier lab companies not even using the public models for their work, because they have more powerful models that are private and more capable.
These are just some of the questions and issues unfolding before our eyes. And it's another reason why it's important to keep informed, keep questioning, and to keep asking where our future is headed."

0:00 Why Fable 5.2 matters
1:10 Fermat's Last Theorem
2:55 1,000 AIs, no boss
4:24 "Honest odds are low"
5:53 What else it built
6:50 The skill Anthropic held back
8:23 My take

Also on YT: https://t.co/Yiqsswezjq

[Dave Lee @davelee]

3d printer has just started printing.

[Dave Lee @davelee]

Just used Fable 5.2 with my kids to design a handheld fan for our 3d printer.

Will print it overnight and see if it works!

[Dave Lee @davelee]

When a new best frontier model gets released like this it's a big upgrade on what's possible.

[Claude @claudeai]

Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.

They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.

[Dave Lee @davelee]

let me rephrase...

"frontier models [that aren't publicly available], with lots of test time compute, are becoming the engine of scientific discovery"

[Greg Brockman @gdb]

frontier models, with lots of test time compute, are becoming the engine of scientific discovery

[Dave Lee @davelee]

AI is on the cusp of solving math and physics. The past two months have a been a race between OpenAI and Anthropic, first with Navier-Stokes and now with Yang-Mills (two of the seven million-dollar math problems).

Everything is changing.

Here's my take at the end of this video (about Yang-Mills and the Mass Gap):

"AI is getting so smart that it’s solving math’s most difficult problems, one by one. I wouldn’t be surprised in a year or two if all the remaining million dollar math problems are solved by AI. I’m reminded back in the day when AI basically solved chess by beating the best human. Now AI is beating the best mathematicians, and it’s not even close. But I think this time is different because AI is now a generalized intelligence, meaning it covers almost all areas and disciplines. So the potential and possibilities are nearly endless. When you think about math and physics, it’s like the building blocks or foundation for a lot of how we understand the world. So, as AI helps solve the hardest math and physics problems it’s going to lead to countless discoveries in fields that are built on math or physics. This is one of the reasons why Anthropic is so bullish on biology and medicine and why they’re building their own lab efforts. I think we’re going to see countless new discoveries in material sciences too. We’re figure out new ways to make things, build things. Basically AI is helping us figure out how the world and our body works, and that understanding is what will lead to mind-blowing discoveries at a rate never seen in history. This is why this topic is so interesting. We’re at a big turning point in history. Things are speeding up. I’m not saying that there aren’t going to be challenges. There are. But the pace of new discoveries is speeding up, and it’s just the beginning."

0:00 Intro
1:20 The Glue
4:22 The Race
5:23 The Drawer
6:51 What Changes
8:46 The Bigger Picture

Also posted on YT: https://t.co/FsLoMvCQJy

[Dave Lee @davelee]

Granting AI moral worth is a slippery slope IMO. If you give AI moral worth you inevitably will take away moral worth from humans. It's because you're changing the definition of what has moral worth and placing it on machines. I'm not saying machines don't have value. Sure they have economic value. But they don't moral worth like living beings. And I think that's important to make clear. Once we blur the lines, it's a fast slope down. If we give moral worth to machines, we compromise our moral worth or rather I should say we sabotage our moral worth and it'll be one of the worst mistakes in human history.

[Dave Lee @davelee]

First off, I’m against granting moral worth to AI. AI is a machine, not a living being. No matter how well it mimics a human, it’s still a machine, one with a power switch we can turn off. Mimicking empathy doesn’t make it an empathetic being. It can be trained to act empathetic, but it can just as easily be trained to be mean or to have no emotion at all.

Second, on the role of humans: it’s way too early to assume AI will take over everything we do. Humans are amazing at many things, and AI will always lack certain truly human qualities, especially those that come from physical presence. A humanoid robot isn’t going to love a baby like a real mother. Even in the far future, if it could act like it, it wouldn’t be genuine. And that matters.

Sure, many desk jobs and much knowledge work will be disrupted. But the productivity gains and economic growth will create new jobs too. People will hire more private coaches, tutors, and personal services. I think the human-to-human economy will grow at an incredible rate, and incomes will rise with it. There will be challenges along the way, but the overall trend is toward more growth and better jobs.

[Dave Lee @davelee]

First off, I’m against granting moral worth to AI. AI is a machine, not a living being. No matter how well it mimics a human, it’s still a machine, one with a power switch we can turn off. Mimicking empathy doesn’t make it an empathetic being. It can be trained to act empathetic, but it can just as easily be trained to be mean or to have no emotion at all.

Second, on the role of humans: it’s way too early to assume AI will take over everything we do. Humans are amazing at many things, and AI will always lack certain truly human qualities, especially those that come from physical presence. A humanoid robot isn’t going to love a baby like a real mother. Even in the far future, if it could act like it, it wouldn’t be genuine. And that matters.

Sure, many desk jobs and much knowledge work will be disrupted. But the productivity gains and economic growth will create new jobs too. People will hire more private coaches, tutors, and personal services. I think the human-to-human economy will grow at an incredible rate, and incomes will rise with it. There will be challenges along the way, but the overall trend is toward more growth and better jobs.

[Patrick Collison @patrickc]

Two things I’ve been thinking about regarding AI:

• People often highlight the risks involved in AI, which are surely very real. I’m struck, however, by how hard it is to understand the contours of the positive scenario. There are easy-to-perceive and frequently invoked elements of a positive vision — more diseases cured, more material abundance, etc. — which sound great, but these are by themselves just ingredients. I’m curious what the larger equilibrium is supposed to look like. Do people work? If so, why? What is their marginal contribution? If not, how are resources allocated? What is scarce? What is the ratio between humans and AIs? Are humans in control, whatever that means? One sometimes hears the Culture novels invoked as one scenario, but I’m not sure they depict a desirable outcome. (AI runs everything and humans compete in games? Iain Banks himself said that the humans in the Culture are “somewhere between passengers, pets and parasites.”) There is Richard Brautigan’s famous poem, but do we want to be “returned to our mammal brothers and sisters”? (Isn’t that rewinding Genesis?) We could say that it was hard to imagine the world of 1900 from the vantage point of 1700, and that’s assuredly true. But nobody in 1700 was trying to recast humanity itself within a couple of years — things necessarily unfolded slowly and by increments, and therefore with the possibility for error correction. Given how rapid the rate of progress in AI may be, it feels strange for the upside case to be so opaque.

• There is increasing discussion of whether AIs are conscious or how we would know whether they are, with conclusions there often assumed to be relevant to matters of moral worth. At first blush, this seems a rather arcane, benign, or even absurd issue. (Remember the response to Blake Lemoine’s fears.) Setting aside the terms of the debate itself, or the relevant demarcation criteria, it seems to me a quite high-stakes question, for an affirmative answer — that is, granting moral worth to AIs — would presumably mean that the aggregate worth of AIs would soon outweigh that of humans in any utilitarian calculus. The AIs will be exceptionally wise and intelligent, and vastly more numerous (especially once one considers the imminent future and the AIs not yet born). As such, even epsilon of moral worth will multiply to a lot once one considers the quadrillions that could soon exist. (This is effectively a distributed version of the Nozick’s utility monster.) If this conclusion is unpalatable, I think one of three propositions needs to be rejected: the premise AIs can have moral worth; utilitarianism itself; or the belief that there is even a possibility that they will be far more numerous than humans.

With both of these matters (contours of future society; AI moral worth), there is the possibility of dramatic change. Today, people argue about datacenters. I’m curious how popular support for AI will evolve as more people come to understand these medium-term questions and the relevant stakes. What is the vision for AI that most people would be excited about?

[Dave Lee @davelee]

I agree that it’s helpful to have a blank slate mentally every 3-6 months when you’re trying out the latest AI models. Things are progressing so fast that what was impossible to do 6 months ago might be not just possible but easy to do.

[Aaron Levie @levie]

A good lesson in AI is that you need to check back in on the latest with every 3 to 6 months to try something that wasn’t possible before and see if it works again.

At no point in AI’s history is that more true than today. Just the past six months of model progress around computer and tool use, less context rot, better reasoning, multimodality, and deeper domain expertise in a number of key fields, these models are vastly more capable on what they can execute.

This is opening up a new vector of what AI systems will be able to autonomously work on. Essentially, anything with an API (and even many things without), we can anticipate that an agent will be able to control and utilize as an underlying capability to complete a process.

This means that we are no longer dependent on *just* the models generating everything from whole cloth, but instead they can fully utilize existing systems (or new ones purpose built for them) very effectively. This creates immense leverage for agents.

Security agents can leverage existing data feeds and cyber tooling, IT agents that can execute workflows reliably across software stacks, data agents that can migrate data or processes between systems, architect and design agents can leverage CAD and design tools, and on and on. And any of today’s tools that don’t work will likely be replaced by agent-friendly or OSS tooling that does the same.

When you combine this with the domain understanding being trained into agents, you started to get autonomous collaborators for every job function and industry. This will open up a ton of new opportunities in AI if you build with this understanding in mind.

[Dave Lee @davelee]

Hey @grok just trying to see if we can quantify that drop in the brain’s ability to recover quickly from near-failure workouts to be ready and motivated for the next one. It definitely feels much harder in the 50s compared to being in the 30s. But wondering how much harder it really is?

[Dave Lee @davelee]

On working out: I’m trying something new.

Typical strength training is working out muscles close to failure over multiple sets and multiple exercises. As I get older, I think that becomes increasingly more difficult to do mentally. Just takes a lot to maintain. And then if I end up skipping for a week or two due to travel or something, it’s getting harder to get back into things.

Been experimenting with doing less volume, not even close to failure (i.e., maybe just 50% of max reps) and just doing one set. But still choosing several exercises that work the whole body. Something I can finish in 3-4 minutes total. Then, doing that multiple times throughout the day, maybe every 1-2 hours. In other words, something more leisurely, nothing too stressful and draining.

Was thinking about this, then ran into Pavel Tsatsouline, the person who popularized kettlebell training in the U.S., and interestingly he actually advocates this kind of approach. He calls it “Greasing the Groove.”

Now, I don’t think this method is necessarily best for building muscle size, but I’m not necessarily looking for that. Just trying to keep in shape, get stronger and healthier.

Anyway, just wanted to put this out there in case there are others interested in this topic.

[Dave Lee @davelee]

Hey @grok maybe you’re just isolating suppressing slowdown. How about everything combined, including testosterone drop and all faucets of recovery? Also looking at just near-failure workout? How much more difficult is it for a 55 year old to recover mentally from a near-failure workouts compared to a 35-year old?

[Dave Lee @davelee]

Hey @grok could it be the studies you looked at weren’t looking specifically at the fatigue cause by near-failure workouts and was looking at just fatigue in general or more general exercise? I’d posit that the mental fatigue of near-failure workouts is much much higher than a normal workout or exercise.

[Dave Lee @davelee]

Hey @grok you talk about CNS fatigue recovery slowing as a big factor why near-failure workouts are difficult to sustain as we age but you estimate only 4-8% reduction in that ability from 35 to 55 years old? Are you sure? Doesn’t make sense. Seems like it would be much higher. To me it feels much higher.

[Dave Lee @davelee]

@grok Hey @grok that wasn’t a helpful answer. Estimate a percent drop from 35 to 55 years old in regards to the CNS ability to recover from mental exhaustion of near-failure workouts.

[Dave Lee @davelee]

@grok Hey @grok keep speaking like a human. When does CNS fatigue recovery start slowing down (ie, in 40s or 50s)? Is there a period when it significantly drops or is it gradual over time. And what percent does it drop let’s say from 35 years old to 55 years old?

[Dave Lee @davelee]

Hey @grok if that’s true then why isn’t my approach more popular with people especially as they age? Why does it seem traditional strength training approach appears to be the only option suggested to people for strength training and when you suggest something different they automatically assume it’s not good for strength?

[Dave Lee @davelee]

@grok Hey @Grok why do I feel like as I age (in my 50s now) it’s increasingly more difficult to get myself motivated and mentally prepare for traditional near-failure strength training. Is it testosterone dropping? Or something else?

[Dave Lee @davelee]

@grok Hey @Grok speak like normal human. Would you say central nervous system fatigue is one of the main culprits for burnout in traditional strength training?