We pitted LLMs against themselves in different languages across games🎮 Usually when a LLM is worse at a task in some language, we go "it's a language issue". But isn't that circular? If the task is made of language, how do you tell if the model lacks skills, or just struggles to read the question? 🧵

Does anyone in my network know when one should buy B6000 vs B200? Or who to ask?

How do projects from Academia become sustainable? You've got a project, and it succeeds, then you've got technical debt and people that care. What's your options next?

Have a complaint about how arXiv works? ArXiv is ready to listen, open community for open science: www.reddit.com/r/arXivMeta/

While language is obviously solved, Google Home can't understand a word I speak to it. Text or speech to text language and cultural issues are countless. This calls for you to join, to co author and to show those model providers they should care about your language and culture as well.

Counterfactual futures are the holy Grail of, well of a lot, right? It's predicting the future plus making A/B tests. Humanities come with this question now. Can we now use AI to reimagine history and see how it would unfold differently?

AI language acquisition. Geniuses debated how humans learn language, what their stimulus is, and what kind of rules do they apply. Why is the way AI learns language, less legitimate field of study? I want the theory of AI language acquisition

I've just got hundreds of new citations in Google Scholar Apparently, you can search your name and first letter + last name and find tons of badly parsed versions of your papers. (Yes, citations don't matter, but making camera ready is boring, and I needed an escape,...)

At last, a way to unlock mulitlingual knowledge sharing in LLMs! 🌍 ​By pretraining an English/Arabic model and swapping Arabic for a word-wise, 1-to-1 translation mapping to English, we saw a massive boost in cross-lingual knowledge transfer 🚀

What do you think ICL changes?

If you reimagined how AI science should be done in Academia, what would you change? I keep posing this question to people I meet lately (and I am going to act on the answers), and thought, there are so many people I don't physically meet...

🚀 I'm delighted to share an **open postdoc position** hosted jointly between our group at @csail.mit.edu and the MIT-IBM Computing Research Lab, mentored by me, Ankit Gupta, @lchoshen.bsky.social, and friends; the start date can be as early as fall 2026.

How can we know more relevant (old?) papers to our work? If before publishing we would have made a call for related work for example? Big ambitious ideas? Surprising tools?

Put bluntly: the reason academics think LLMs are left-wing biased is that LLMs think academics are left-wing biased. As the auditor tries to infer the politics of the model, the model is simultaneously inferring the politics of the auditor.

How can people not love working in the open? All crunching towards Neurips, then a collaborator just pops, hey I did a "quick" another section for the paper where it was a bit weak, cool everyone?

LLMs can't generalize across 2 copies of the same language!🫠 Usually we blame multilinguality issues on syntax, tokenizer fragmentation or data disproportion, so what if we eliminate these factors? In our ongoing work by Adam Jaber Bobby Cheng and I we ran a pretraining experiment doing just that

Created a minimal skill to search for grants, I assume people around have something better? Or is it worth making a git out of it and we will start improving on it together? github.com/borgr/grant_...

Chinese babyLM is on folks:

Money💶, academic money🧑‍🏫 to be exact, is almost a mystery🕵️ GPU hours, hidden philanthropies, industry websites, governmental lesser-known funds. How do you know who wants to fund your research? Starting a thread for hidden grants, locations to find them, and ideas for a change

Alignment to animals 🐄🐖🐓 Just read a fun paper training\eval models to care about animal welfare. On one hand, it shows we can plug values into these systems. On the other… A weirder question: what values are we absorbing as we happily suckle on AI-generated words? 🤖🥛

By MIT and IBM Why inference scale when you can scale scale? Our thinking about thinking changes scaling completely! By asking the model to think about what it would have thought for N steps It breaks benchmark after benchmark Games, Scientific discovery, proteins... #AI

Rebuttal season is here, yey🤞 With many asking me, I compiled the most common misconceptions Hope the tips help 🧵 All tips: docs.google.com/document/d/14Wax8M5w8F_8miDlYJ9-I6wqpelxlXjCEUbkNzNMqqE/edit?tab=t.0#heading=h.rfq27f356vmm #AI 🤖📈🧠

Do you run pairwise evaluation? Do you test your models on the Arena-Hard and AlpacaEval benchmarks? You probably want to read this 🧵👇 arxiv.org/abs/2603.16848 By Shachar Don Yehia, me and Omri Abend 🤖📈🧠 #AI

You ask in one language and expect in the other, models would fail to provide... They predict when models will succeed (👇) They also show the non-shared entities hold some of the problem, thinking on the right one helps, but is far from explaining the gap. #AI #LLMs arxiv.org/abs/2603.17070

When you restart your embeddings language transfer is better If you restart embeddings in pretraining, then in continual pretraining for a new language, you get more language transfer.

Evaluation research? No place like evalEval

This got awfully little reaction, so there's a clear agreement this is our expectation from the future of the field?

New army🪖 dataset asserting that LLMs answer on violence, terrorism, and war topics Military experts manually created challenges. To their dismay, models often did not answer. Until they steered them It is definitely novel. What do you say, wonderful research? Horrible? Why?

Interesting concept. Feeding assistant turns back into the model isn't always necessary and may be context pollution. However, I'd hesitate on this considering a lot of use cases are about iterating and changing only a small portion of prior output (which sometimes fails with current models).

Do LLMs Benefit from Their Own Words?🤔 In multi-turn chats, models are typically given their own past responses as context. But do their own words always help… Or are they more often a waste of compute and a distraction? 🧵 arxiv.org/abs/2602.24287

👩‍💻 Running evals? Contribute your evaluation data to the community! Add your data to Every Eval Ever as part of the ACL Shared Task evalevalai.com/events/share... (Any eval data welcome - and we provide converters from lm-eval, Inspect, and HELM) Submission deadline: May 1, 2026

👶 BabyLM is back at EMNLP 2026! We are excited to announce that the 4th BabyLM Challenge & Workshop will once again bring together researchers interested in sample-efficient, developmentally plausible language modeling. @emnlpmeeting More in🧵

Agents should be general Why are we building code agents, CLI agents, browser agents separately? Why does adapting to a new benchmark take a month? Our collaboration brings diverse views, pros here cons in the paper & Your push back if I’m wrong Argument + paper link👇🧵 #MCP #ai #LLMs #agents 🤖📈🧠

Don't complain. Do it yourself. When the @eval-eval.bsky.social coalition started studying together what is broken in evaluation, I knew what we need to do. We need to digitize evals. How come every evaluation is reported differently? In a separate place? We built Every Eval Ever: 💪🤖📈🧠

LLM "robustness" is often treated like a mysterious, standalone capability. But what if it’s not? 🤔 Our new research shows robustness naturally appears when models truly understand a task - competence drives stability. More details in the thread 👇 🤖📈🧠 #AI arxiv.org/pdf/2602.03344

I just saw an LLM making a fluency mistake. How can that happen?! Something about long context?

"Kill your darlings, cut unless serving the story" One tip from the guide. If you needed a moment out of #ICML2026 writing and graphs, why not read some writing and figure making tips? docs.google.com/document/d/1...

"It is time to separate language from language models" The revelation keeps bugging me, so it also underlies the talk I just gave "multilingual?" Thought I'd briefly share the contents of the talk: 🤖📈🧠

From the conference that brought you the AC instead of rebuttal: Desk rejecting months after reviews were given. @iclr-conf.bsky.social - initiatives are good, but...

It'd be interesting to study how the substance of what functioning robots learn differs from LLMs. LLMs are based on language which in a sense is at least one level of abstraction up from experiential reality; language is humans' expression (a compression) of their collective experiences.

On the unexplained similarity across networks In behavior, order and weights, we keep seeing evidence that learning is more consistent than one might think. A walk through occurrences, my thoughts and the open question, why?! What's your hypothesis, missed papers and thoughts 🤖📈🧠 #AI

“Today, I have a vision, a vision of superintelligence from experience” Presented in his humble way, Rich Sutton shares his vision of what AI needs General, experiential, discovers its own abstractions and not bitter🤢 #NeurIPS2025 #NeurIPS 🤖📈🧠

LLMs do not learn from experience LLMs do not learn from explicit corrections LLMs do not learn from being told the answer LLMs do not learn from being shown how to solve it We study Machine Learning, these are opportunities! A gold mine of research.

"Hey dude, look. What is this button doing?" ⚡️BzZzZz⚡️ "Hey dude,..." Would you press the button again? Would an LLM? Evolving LLMs, diverse open LLMs, and their evaluation are on my mind. Before I share more, I encourage you to say hi here or in #NeurIPS 🤖📈🧠

Join us at NeurIPS 2025 for the MindGames Challenge Workshop! Explore theory of mind, game intelligence, and multi-agent LLMs in interactive game environments. 🗓 Sunday, December 7 ⏰ 8:00–10:45 AM 📍 San Diego Convention Center, Ballroom 6CF 🤖📈🧠

Quietly quietly Yuntian Deng released 2 million new chats of real humans with models (non-toxic), Kudo. There are now datasets of over 4.5m chats open for research and all in the same format (shareLM)! huggingface.co/datasets/sha... t\h @msheshera.bsky.social

001-110 101-? AAT-TTA TAT-? In context learning emerge outside language wonderful finding

The Golden pacifiers are ready See you soon in BabyLM (emnlp)

Introducing Global PIQA, a new multilingual benchmark for 100+ languages. This benchmark is the outcome of this year’s MRL shared task, in collaboration with 300+ researchers from 65 countries. This dataset evaluates physical commonsense reasoning in culturally relevant contexts.