Jiaming Zhang

RUC STAT｜Intern @UofIllinois @KAUST_NEWS｜Trustworthy LLM & Deep learning theory

[Jiaming Zhang @traderbxy]

@noahatwhere 真实…

[Jiaming Zhang @traderbxy]

@letyrodri That may be a misunderstanding. I met with nitpicking.

[Jiaming Zhang @traderbxy]

Got a 2/2/3 on my paper. Since GPT-6, it feels like every reviewer has turned into a math genius, precisely attacking my assumptions and constructing increasingly sophisticated numerical counterexamples.

[Jiaming Zhang @traderbxy]

@DireDoge1029 But human judgments of novelty don’t follow any fixed standards either. They can be highly subjective and inconsistent, potentially even less reliable than evaluations based on a well-defined set of criteria.

[Jiaming Zhang @traderbxy]

@Lptomov82 Yes, but it is hard to evaluate in current review criteria

[Jiaming Zhang @traderbxy]

And if research itself is becoming increasingly automated, what’s left for humans to do?

[Jiaming Zhang @traderbxy]

Do we still need conferences? Do we even need human peer reviewers?

Could we develop something like Lean-style formal verification for research papers, verifying validity, novelty, etc., and accepting papers based on verifiable guarantees rather than noisy human judgments?

[Owen Queen @oq_35]

1/n Within 15 months of publication, most of the LLM claims we tested either no longer held or showed substantially diminished effects. How long does our knowledge of AI hold up? In our new study, we revisit 50 findings to see what fades, what persists, and what gets stronger. 🧵

[Jiaming Zhang @traderbxy]

mark

[Yuejie Chi @yuejiec]

Check out our book on statistical foundations of RL! It took us quite a while, but procrastination allows us to use AI at the final stage to polish a few things. We appreciate your feedback.

[Jiaming Zhang @traderbxy]

that’s insane

[Dan Roberts @danintheory]

We are announcing three new discoveries in mathematics, computer science, and physics about harmonic analysis, complexity theory, and quantum field theory.

https://t.co/9BX8HWQDTJ

[Jiaming Zhang @traderbxy]

If expectations aren’t met, the fallout from capital markets could be brutal.

Maybe it’s time for us to slow down, consolidate what we’ve achieved, and focus on delivering real results.

[Jiaming Zhang @traderbxy]

Scary to think about, but it might be true.

I feel like OpenAI has pulled too much optimism forward—announcing too many results before they were fully mature and raising enormous amounts of capital on those expectations.

The media can hype things up, but reality doesn’t care.

[Jiaming Zhang @traderbxy]

@elonmusk @Yuchenj_UW Scary to think about, but it might be true.

I feel like OpenAI has pulled too much optimism forward—announcing too many results before they were fully mature and raising enormous amounts of capital on those expectations.

The media can hype things up, but reality doesn’t care.

[Jiaming Zhang @traderbxy]

Actually, I have provide a similar conclusion in my previous paper published on ICML2026 from a feature learning perspective . The attention or softmax is an Adaptive-Weight MLP. It is used to accelerate convergence as the weight of signal is growing quickly.

[Pedro Domingos @pmddomingos]

TL;DR: An attention head is a fast-weight MLP.

[Jiaming Zhang @traderbxy]

First time reviewing for ICLR. Spent an entire day carefully reading three papers, trying my best to write fair, thoughtful reviews that AI could NEVER have generated.@iclr_conf

[Jiaming Zhang @traderbxy]

optimal transport is sooooo difficult

[Jiaming Zhang @traderbxy]

catch up with many friends and advisors, and also make lots of new friends. #ICML2026

[Weijie Su @weijie444]

We released "The Newton--Muon Optimizer" . We show that Muon is secretly an implicit Newton method, and use this insight to build a better one. 1/n

Paper: https://t.co/Ua54426bWB