Tom Backert
525 posts
Tom Backert
@tb18_08
Computer Scientist
San Francisco, CA Katılım Aralık 2024
405 Takip Edilen106 Takipçiler
If you're curious how I managed to do over $20,000 in inference on the last 48 hours, here's a video all about it.
Spoiler: loops are really powerful
English
@bcherny This is the move to only understand the code on a meta level from now on. Understand the big picture but don’t bother with the details.
English
I've been using Artifacts in Claude Code for everything: visual explanations of tricky code, system diagrams, quick previews of a few animation options, data analyses and dashboards I share with the team. They are a game changer for how I work with Claude. Can't wait to hear what you think!
Claude@claudeai
New in Claude Code: Artifacts. Interactive pages built from your session, like a PR walkthrough or a living project dashboard, shared with your team at a private link. Available in beta on Team and Enterprise plans.
English
Tom Backert retweetledi
Tom Backert retweetledi
Brainrot doesn't just steal your time. It trains you to hate the things that would actually save you.
English
At this point, parents should just forbid social media. We are priming future generations for loss by showing them how strangers “succeed.”
kache@yacineMTB
I wonder if 17 year olds these days have the attention span to peel an orange, after all of the cognitive abuse at the hands of neural recommendation addiction optimizers
English
@yacineMTB @MParakhin Novel ideas separate from average but scaled ideas
English
Model training is a game where GPUs and data are an overwhelming advantage. So when Recraft beats xAI, DeepSeek, Meta, BFL, Microsoft, etc. with a tiny fraction of the resources, the conclusion is: big-company ML talent selection is broken. Very different from "AI experts" :-)
English
Writing very incisive and reflective commit messages that nail down the changes made lets you stay longer in the loop.
English
@mentalgeorge Takeaway: Anything that can be done can be understood, and anything that can be understood can be done.
English
1. Dwarkesh is right: intelligence is a weak predictor of power in humans.
2. Instead, close study of powerful humans suggests that power is downstream of properties like charisma, unscrupulousness, and ambition.
3. But is this likely to be true of the AI superintelligence? How might an intelligence-maximalist respond to Dwarkesh? I think they could say something like:
"Sure, powerful humans are more likely to be charismatic than smart. But that's fundamentally a skill issue for humans, not a fact about intelligence per se.
Trump can outplay a physicist at social-manipulation only because the physicist *isn't quite smart enough* to properly model the social interactions in question. A true superintelligence would!"
3. In short, the intelligence-maximalist can claim that Trump's strategy for acquiring power is akin to the way some birds use magnetic sensing to navigate the earth - a computationally efficient "hack" that could, ultimately, be instantiated through intelligence alone.
4. So, although real-world-Trump convinces millions to vote for him through his unbeatable mastery of Magic Charisma Sauce, even Magic Charisma Sauce can be trivially outmogged by the unbridled IQ of a superintelligent TrumpGPT.
5. Claiming that, in the limit, intelligence == power therefore looks like a safe bet: it is a bet that magic does not exist. Anything that can be done can be understood, and anything that can be understood can be done. Anything the charismacels can do, the superintelligence can do better.
6. But I'm not so sure. Magic may exist after all. It's unclear to me that the world is in fact infinitely modellable and computationally reducible, particularly by one entity. Dynomight's "limits of smart" is the best I've read on the subject - the superintelligence may still lose to Stockfish at chess and/or fail to predict the weather.
(there is, of course, an entirely different objection to Dwarkesh's intelligence =/= power, which asserts that this non-equivalence is a contingent outcome of the fact that humans are bottlenecked by things like desire and ambition, none of which would apparently impinge upon the silicon supergod)
Dwarkesh Patel@dwarkesh_sp
# The mistake of conflating intelligence and power I had an interesting discussion recently. Someone asked me, what is intelligence? I said, the ability to achieve your goals across a wide range of domains. Okay, he says, then by that definition isn’t Donald Trump the intelligent person in the world, followed in quick succession by Xi Jinping and Vladimir Putin? To be clear, these people are obviously very competent and clever. But when you think of ASI, you don’t think of Trump, but more so. The person who kept pressing this question was correctly pointing out that I basically defined intelligence as power. And by this definition, Stalin was the most intelligent person who ever lived. Now, of course, you could change the definition of intelligence to something more like, manipulate abstract concepts and rotate shapes. But notice that the most powerful people in the world do not max out this quantity. The correlation between extreme power and this kind of intelligence might be even weaker than the correlation between extreme power and height. The physicists are not running the world. We tend to conflate power-seeking AI and superintelligent (in science and tech) AI. I’m not denying that AI can be power-seeking. Whatever skills and drives Donald Trump has could be embodied in a digital mind. I’m simply pointing out that the way AI systems are currently becoming smarter (by getting trained to be to be really good at specific economically valuable tasks like coding) is not that strongly correlated with power. We often talk about power in this way that misunderstands how it is actually derived in our world. Our intuitions are primed by games like Diplomacy or Go, which are designed to isolate and reward a g loaded kind of strategic reasoning. But in the real world, power is more the product of having the authority and trust to get lots of people to collaborate with you, rather than some galaxy brain scheming capability. Trump is not powerful because his brain, considered in isolation, is the most effective optimization engine on Earth. He is powerful because the government which hundreds of millions of people consider legitimate gives him a lot of authority. A group versus individual level analysis is useful here. As @GarettJones has written a lot about, individual IQ is only modestly correlated with individual income, but national IQ is strongly correlated with national outcomes. This is because intelligence has a lot of spillover effects - smarter societies cooperate more, save more, and can coordinate to build things like space shuttles and semiconductors. Richard Trevithick, who invented the high-pressure steam engine, died in poverty, buried in an unmarked pauper’s grave. But the fact that 18th and 19th century Britain had lots and lots of people like Trevithick contributed to Britain being able to set up a global empire and outcompete lots of backwards principalities around the world. It seems to me that the right mental model is that automated firms will outcompete everyone else in normal capitalist ways, rather than a single AI outthinking everyone else.
English
English
@elonmusk @eastdakota Doing some assembly so others can do some vibing.
English
Yes.
It’s not that we’ve discovered some magic bullet, but rather that JAX, or at least the open source version of it, is mostly optimized for small to medium-sized training runs on Google TPUs, whereas we need to massive training runs on Nvidia GPUs.
Pipeline parallelism is essential and crushes fully-sharded data parallelism at scale.
And C will compile to the most efficient binary short of assembly. Maybe we will do a little assembly too.
English
There’s a ton of headroom to better optimize AI training and inference.
Elon Musk@elonmusk
SpaceX has almost finished writing V1.0 of an in-house AI training stack in C that exact-maps to 220k GB300s with 800G NICs, making heavy use of pipeline parallelism and getting as close to bare metal as possible. The potential speed improvement vs JAX for large training runs is over an order of magnitude.
English
Tom Backert retweetledi
Being busy is not the same as being productive.
The 80/20 principle, also known as Pareto’s Law, dictates that 80% of your desired outcomes are the result of 20% of your activities or inputs.
Once per week, stop putting out fires for an afternoon and run the numbers to ensure you’re placing effort in high-yield areas:
What 20% of customers/products/regions are producing 80% of the profit?
What are the factors that could account for this?
Invest in duplicating your few strong areas instead of fixing all of your weaknesses.
English
A picture is worth a thousand words.
Why does X make me think more than a thousand Instagrams then?
English