The Biology of a Large Language Model
We investigate the internal mechanisms used by Claude 3.5 Haiku — Anthropic's lightweight production model — in a variety of contexts, using our circuit tracing methodology.
Posts that commenters keep linking to, ranked by how many different people have cited them in Hacker News comments. 33 posts.
We investigate the internal mechanisms used by Claude 3.5 Haiku — Anthropic's lightweight production model — in a variety of contexts, using our circuit tracing methodology.
There is a rush for AI companies to team up with space launch/satellite companies to build datacenters in space. TL;DR: It's not going to work.
Research on the ability of large language models to introspect on their own internal states.
What do Americans die from, and what do the New York Times, Washington Post, and Fox News report on?
Research from Anthropic on the faithfulness of AI models' chain-of-thought reasoning.
A guest post by David Schneider-Joseph
Craig Gidney's computer science blog
Batteries are now cheap enough to unleash solar energy’s full potential, getting as close as 97% of the way to delivering constant electricity supply 24 hours across 365 days cost-effectively in the sunniest places.
So, have a seat. Put your feet up. This may take some time. Can I get you some tea? Earl Grey? You got it.
SESSION 7: SPECIAL SESSION, 17th Annual EOS/ESD Symposium THURSDAY, SEPTEMBER 14, 1995, 8:00 am SESSION 7: SPECIAL SESSION: ELECTROSTATIC CONSIDERATIONS IN INDUSTRY MODERATOR: D. Swenson, 3M 7.7 CASE STUDY - LARGE PLASTIC WEB ELECTROSTATIC PROBLEMS, RESULTS, AND CURE, D. Swenson, 3M Company Tremendous static charge…
In this post I explain why today’s humanoid robots will not learn how to be dexterous despite the hundreds of millions, or perhaps many billions of dollars, being donated by VCs and major tech companies to pay for their training.
Before we get nerd sniped by the shiny engineering details, ask the only question that matters. Why compute in orbit? Why should a watt or a flop 250 miles up be more valuable than one on the surface? What advantage justifies moving something as mundane as matrix multiplication into LEO?
Dolphin researchers are using Gemma and Google Pixel phones to try to decipher how dolphins talk to one another.
Physicist Francis Perrin sat at a nuclearfuel-processing plant down in the south of France, thinking to himself: “This cannot be possible.” It was 1972.
“From a mathematics and trajectory standpoint and with a certain kind of technology, there’s not too many different ways to go to Mars. It’s been pretty well figured out. You can adjust the decimal places here and there, but basically if you're talking about chemical rockets, there's a certain way you're going to go…
INTRODUCTION TO DIGITAL FILTERS WITH AUDIO APPLICATIONS
Can optical illusions take you outside the human color gamut?
Bending Spacetime in the Basement
Open Access Publication & ACM
Water infrastructure often gets less attention and focus than other types of infrastructure.
Libraries have been trying to collect humanity’s knowledge almost since the invention of writing. In the digital age, it might actually be possible to create a comprehensive collection of all human writing that meets certain criteria. That’s what shadow libraries do - collect and share as many
I believe that language models aren’t world models. It’s a weak claim — I’m not saying they’re useless, or that we’re done milking them. It’s also a fuzzy-sounding claim — with its trillion weights, who can prove that there’s something an LLM isn't a model of? But I hope to make my claim clear and persuasive enough…
For overdetermined reasons, I’ve lately found the world an increasingly terrifying and depressing place. It’s gotten harder and harder to concentrate on research, or even popular scienc…
Marble, our frontier multimodal world model, is available to everyone starting today
Anthology of sociology, statistical, or psychological papers discussing the observation that all real-world variables have non-zero correlations and the implications for statistical theory such as ‘null hypothesis testing’.
I shared a controversial take the other day at an event and I decided to write it down in a longer format: I’m afraid AI won't give us a compressed 21st century.
[You can follow me on social media: @rodneyabrooks.bsky.social and see my publications etc., at https://people.csail.mit.edu/brooks ]
How early, sustained, supermassive black hole jets carved out cosmic voids, shaped filaments, and generated magnetic fields