my thoughts and prayers with whoever at openai is responsible for making good use of these chips

this is what good abstraction looks like. seems so simple in hindsight and yet, you know a lot of thought went into it. can't wait to try it out!

first we shape our tools. then our tools shape us.

State of the art tool use capabilities. Let's go!!!

"Of course it tops the leaderboards" :)

Postcards from London

Massive jump in tool use capabilities at a fraction of the price! Onwards and upwards ⚡⚡⚡

Truly a flash-ship model 😀

no question about it

Heisenberg’s Uncertainty Principle (Home Edition): You can’t have a clean desk and a clean drawer at the same time.

Anyone for badminton tomorrow (Saturday) from 10-11 AM in Bellandur?

rohan anil I think a lot of good agentic harnesses (eg coding) are highly tuned for specialized use cases, which make it significantly better than sum of its parts. Viewing it purely from the lens of model building misses out the complexities in the harness which actually make it awesome.

rohan anil Building useful agentic harnesses is hard and severely underrated IMO. It requires a deep understanding of model behavior, often not captured in evals. A lot of people building models are far too comfortable just looking at numbers instead of deeper analysis.

Hi Everyone, we are hiring for Ph.D student researchers in the field of ``Search/RL with LLMs'' ideally for discovery. Please respond in this form if you are interested. Please don't reach out directly as i may not be able to reply individually forms.gle/CxY4VzQRdJacLX…