Yoonho Lee
@yoonholeee
Final-year ML PhD @StanfordAILab + SR at Google.
Trying to make text optimization work.
665 Following 5K Followers
Harness optimization is sample-efficient but plateaus. What should you do if you can afford to update the model too?
Introducing WHALE: a simple recipe for jointly optimizing an LLM's weights and harness.
Blog:
Paper:
Show more
Please sign up if you've recently worked on a paper!
By our estimates, there are at least 10,000 eligible authors worldwide. We’ve reached roughly 1% of them so far. Help us reach the other 99%!
There’s a time-sensitive aspect to this. We want to capture fresh research insights before they become common knowledge (e.g. pretraining data)
I also think the resulting dataset could be a great resource for human researchers. We rarely get to see how other groups decide what is worth building on. Everything will be released publicly.
Read @ohmyksh 's thread below for more details
Show more
How can we autonomously improve LLM harnesses on problems humans are actively working on?
Doing so requires solving a hard, long-horizon credit-assignment problem over all prior code, traces, and scores.
Announcing Meta-Harness: a method for optimizing harnesses end-to-end
Show more
