Justin Kang

PhD Candidate @berkeley_ai; prev Student Researcher @google

[Justin Kang @Justinkangs]

@DimitrisPapail Also having very little luck getting Fable to do basically any work remotely related to security or safety

[Justin Kang @Justinkangs]

@gdb Why are you telling us? Go make it better

[Justin Kang @Justinkangs]

@jasondeanlee @roydanroy Been having tons of issues too

[Justin Kang @Justinkangs]

@melissapan @abeirami @AlexGDimakis Great discussion!

[Melissa Pan @melissapan]

The fun continues 🔥

Now, we have our second panel with: @abeirami @AlexGDimakis and Mohammad Alizadeh

Come by to hear more about hot takes from the field’s thought leaders!

[Hongxun Wu @HongxunWu]

🧵(1/8) An @OpenAI internal reasoning LLM achieved an AI Math milestone: solving an open problem central to its mathematical subfield— in this case, the unit distance problem of discrete geometry.

We came across it in a side quest to truly push our model on the hardest problems.

[Justin Kang @Justinkangs]

The legend @PGelsinger has been vindicated!

[Justin Kang @Justinkangs]

I think the most beautiful application of information theory is "achievability" proofs. These prove that some engineering tasks is theoretically possible. Humans are an achievability proof for current AI. As we move beyond human capabilities this is no longer the case!

[Dimitris Papailiopoulos @DimitrisPapail]

Information theory is a mathematical theory and a language, not a scientific theory that explains phenomena. Slapping mutual information on all of AI’s mysteries won’t help explain them.

[(((ل()(ل() 'yoav))))👾 @yoavgo]

even if we do consider math as the pinnacle of human intelligence, it is not the kind of math that proves stuff, but the kinds of math that invents new concepts and then invents questions about them

[Justin Kang @Justinkangs]

@yoavgo Very good point!

[Justin Kang @Justinkangs]

@aminkarbasi This monograph is a great reference. I referred to it often when I was doing my work on formalizing some deep connections of group testing to data/feature attribution during my PhD.

[Logan Kilpatrick @OfficialLoganK]

Very excited for all the stuff the @antigravity team has been cooking :)

[Samuel Marks @saprmarks]

Surprising results that give a great reminder we don't really know why and when synthetic document fine-tuning works to modify LLM beliefs.

[Harry Mayne @HarryMayne5]

New paper:
We create documents describing ridiculous claims (e.g. Ed Sheeran winning the 100m gold medal) but surrounded by warnings that they are entirely false.

When we finetune models on these documents, they end up believing the claims are true!

[Suraj Srinivas @Suuraj]

"We think: somewhere inside the model there is a clean, human-legible structure waiting to be recovered"

This core philosophical belief separates "believers" from interp skeptics.

But with neural networks the lesson seems to be: you get what you optimize for, nothing more.

[Harry Mayne @HarryMayne5]

To be presented at ICML! 🇰🇷

[Kangwook Lee @Kangwook_Lee]

https://t.co/BSC1aLHwps

[Justin Kang @Justinkangs]

@thegautamkamath That is really cool! I am writing a BITS article, how did you sign up to get a physical copy?

[Sara Hooker @sarahookr]

A really excellent book.

A few people independently told me this was one of their favorite books over a decade ago. I bought it, and it became one of the textbooks on my shelf I revisit from time to time to spark the joy of holding ideas to a different light.

It brings to life the elegance of information theory.

A good day to recognize 10 years from the passing of David McKay.

[clem 🤗 @ClementDelangue]

We need more open agent traces datasets. Who can help?