yusufozkan

"We gave Claude a prompt to search through a massive database of DNA sequences for interesting new examples of RTs. Our involvement was limited to the initial prompt and the lab work, while Claude agents combed through the database, investigated the distinct RT families, and used their own judgement to identify interesting candidates. After 21 hours spent searching this data by roughly 950 agents using 210 million tokens, one of the agents spotted something remarkable: a repeating pattern of DNA sequences that occurs next to the gene for an odd-looking RT. After further analysis and testing in our lab, we recognized that this pattern marked a previously uncharacterized enzyme system found in bacteriophages (the viruses that infect bacteria) that we call array-associated reverse transcriptases (ART)."

what the hell is that score/cost curve lol

> arc-agi-3 30.2%

wow

What is your definition of "actually intelligent"? I believe LLM's are more intelligent than the average human in a lot of ways according to the Legg/Hutter definition of intelligence: "Intelligence measures an agent's ability to achieve goals in a wide range of environments".

"This is a general-purpose LLM. It wasn’t targeted at this problem or even at mathematics. Also, it’s not a scaffold. We have not pushed this model to the limit on open problems. Our focus is to get it out quickly so that everyone can use it for themselves." - Noam Brown (OpenAI reasoning researcher) on X

"The proof came from a general-purpose reasoning model, not a system built specifically to solve math problems or this problem in particular, and represents an important milestone for the math and AI communities."

but people here had told me llms just predict the next word

This is the same company that started as a nonprofit dedicated to open AI safety research, then became a capped-profit entity, then effectively closed-source, then dropped the cap, and is now pursuing full for-profit conversion. Every single guardrail they've set for themselves has been quietly revised or removed once it became inconvenient. Anyone want to bet on how long those exclusions last?

> while pre-training our Llama 4 Behemoth model using FP8 and 32K GPUs

I thought they used a lot more GPUs to train frontier models (e.g. xAi training on 100k). Can someone explain why they are using so few?

This is another cool example from their blog

https://imgur.com/a/Svfuuf5

Generating an image of a completely full glass of wine has been one of the popular limitations of image generators, the reason being neural networks struggling to generalise outside of their training data (there are almost no pictures on the internet of a glass "full" of wine). It seems they implemented some reasoning over images to overcome that.

Are you sure you are using the new 4o image generation?

https://imgur.com/a/wGkBa0v