If you're proofreading someone's draft, upload it to whatever LLM you have access to and just chat through the paper. Ask it questions and talk through your confusion. Then send the entire conversation to the author. That's actually good feedback.

Explaining my research to family when they ask what I've been working on

should i open source my cheatsheet

BrownUAlert: Safety and Security on the Brown Campus Given the deeply tragic events of recent days and the status of the ongoing investigation, we understand the fear and anxiety that so many Brown community members are feeling right now. Please be assured that the safety of our

"X is All You Need" title was great but I think we've earned the right to let it rest or just go all in with "All You Need is Whatever I'm Working On" honestly I would respect that more

If you’re a Brown student, locked down in your dorm or the OMAC, and scared and sad and exhausted and angry, know that there are over 100,000 alums around the world who are thinking of you and have your back.

Sorry for not offering this earlier, but if you're a parent of a Brown student and they need help (a ride, a coffee, whatever), please email me and I'm happy to try to help. https://t.co/waNw2eJfzN

Statement on the shooting at Brown University:

BrownUAlert: 8th Update, The Brown campus continues to be in lockdown, and it is imperative that all members of our community remain sheltered in place. This means keeping all doors locked and ensuring no movement across campus. The law enforcement response remains ongoing.

BREAKING: Footage released of Brown University shooting suspect. He's still at large. If you know who he is, call 911

@amaarora Check this out: https://t.co/zsfObLkLhR We recently integrated tool description optimization into GEPA. Regarding expensive calls, the whole point of GEPA is that it is very sample efficient for complex workflows as we showed in the paper. You can reduce the valset size to

was reviewing for finals and got distracted testing gpt-5.2 on a little eval I have. Hmm

Four obvious things that became obvious (or just true) only recently: - Our models are incredible. - Our learning algorithms work. - Our models are unreliable. - Our learning algorithms are wildly inefficient.

@Zoom congrats on SOTA!

Zoom achieved a new state-of-the-art (SOTA) result on Humanity’s Last Exam (HLE): 48.1% — outperforming other AI models with a 2.3% jump over the previous SOTA. ✨ HLE is one of the most rigorous tests in AI, built to measure real expert-level knowledge and deep reasoning across

ICML'26 (abs): 44 days. ICML'26 (paper): 49 days. ECCV'26: 85 days.

RL for reasoning often rely on verifiers — great for math, but tricky for creative writing or open-ended research. Meet RARO: a new paradigm that teaches LLMs to reason via adversarial games instead of verification. No verifiers. No environments. Just demonstrations. 🧵👇

Come join our team - we raised a $9.4M seed round! AI will empower people to do meaningful, high-impact work—yet most people don’t know where to start and are being left behind. We use AI to help businesses uncover their best automation opportunities and build those automations

#KostasKeynoteLessons: Curious about the "Keynote magic" behind my slides? I’m releasing the full Keynote source file for my recent Gaussian Splatting lecture, all 10 GIGAbytes of it! Grab the files in the thread and feel free to remix.