Is it true?

sad

can anyone recommend an AI proof-reading service, please?

MindGames report is out: arxiv.org/abs/2605.29512 cc: @sharky6000.bsky.social do we discuss it on this platform?

I follow like 500 people and only like 10 post anything

cool!

I always wanted to post something overwhelmingly positive and promotional on Bluesky to encourage the corresponding mindset here. However, there's too little positive in my life. Neverthless, today I've bought a Lewis' Caroll book for 15 quid. It's good. Have a a nice weekend and further, everyone.

first page of the pthead book broke me

Of course, I won't publish it fully but I think it has become better since the last year

I sat down to update my research statement — and oh, my god

see, people do a lot of different reinforcement learning methods for LLMs and they barely outperform evolutionary strategies for low-rank matrices, and everybody tries to make it look "okay" lion doesn't concern themselves with obtaining gradients

I have never had a graduation ceremony

It's insanely good perspective on open-ended learning and cultural evolution: inherentlabs.ai Edward Hughes' hand can be seen. Also, I am quite interested how some fancy names and a good pitch can make people invest millions of pounds.

SokoBench: Evaluating Long-Horizon Planning and Reasoning in Large Language Models Sebastiano Monti, Carlo Nicolini, Giovanni Pellegrini, Jacopo Staiano, Bruno Lepri Action editor: Marc Lanctot https://openreview.net/forum?id=pLosAkOoGU #planning #language #reasoning

insane glowup

A cool proxy for critic updates: CoT traces instead of real updates: arxiv.org/abs/2605.04906 Simple and cool idea but - yet - not reliable due to uncontrolled error accumulation

yeah I've been telling people that many MA benchmarks are bollocks thank you for the confirmation p.s I hate smac p.p.s. I hate smac v2

Look at this cookie by Dr Gemp: openreview.net/pdf?id=GlvYT...

I can't proofread my own text. LLMs can't too, I guess...

As we build more cooperative multi-agent envs, we should ask: are we really testing the properties that make Dec-POMDPs hard, partial observability and decentralised coordination, or can agents succeed via shortcuts? 🤔 Our #AAMAS2026 Oral probes this question 🔎🧵

ayo

I reckon we all should try www.ecosia.org

Unfortunately, this works

value-driven transport! a new framework for generative modeling, combining elements of * optimal control / RL * optimal transport * stochastic primal-dual optimization thread about our new work with @pmorenoz.bsky.social (@upf.edu) & Adrian Müller (@ethz.ch) 1/

i would like just to attend RLC

Got a great TMLR paper but missed the RLC deadline? Following last year’s success, @RL_Conference is back with a Journal-to-Conference track! Accepted TMLR papers within scope are invited to submit for consideration. Please submit here: docs.google.com/forms/d/e/1F...

Can anyone share their perspective on how you create research plans and stay aware of what is going in your research field?

Next week I will be at AAMAS looking in awe @kale-ab.bsky.social presenting as an oral our paper on probing actual reasoning capabilities in Cooperative Multi-Agent Reinforcement Learning. More than happy to chat about this project or research in general!

Canadian grand prix this week

bookmarked

Btw, doesn't the fact of the authors uploading a pre-print before anonimously submitting it to a conference eliminate logic of not using AI during the reviews because the reviewer didn't leak anything to the Internet, it was already there... Thus, should deanoning own work imply a desk reject?

Xinlun Zhang, Zhechen Li, Yongzhi Cao, Yu Huang, Hanpin Wang Probabilistic Mechanism Design in Diffusion Auctions https://arxiv.org/abs/2605.17221

PEOPLE LISTEN WHAT IS THE POINT OF ANONYMOUS REVIEW PROCESS IF YOU SUBMIT YOUR PAPER ON ARXIV AND DON'T EVEN ALTER THE TITLE COME ON BE CONSEQUENTIAL PLEASE

Interesting that Google has hired so many people posting things about Gemini so may even trick you into thinking that it's a good model.

Interesting paper: arxiv.org/abs/2605.15331 Information-driven exploration

I want to leave a little comment to all reviewers that ask to enlargen font of my paper's plots. 3 words: "Just zoom in"

multiprocessing vs multithreading in python baarse.substack.com/p/free-threa... not everything is as glaringly simple

Quite an interesting paper. The diagram is telling. However, I am not sure about the critic because you don't really use it in the derivations. Meanwhile, I agree that the optimisation should stay 2-scale, I am not convinced that we should go with not necessary Markovian model and TD updates.

this list is exactly the line that separates me from being called an "intellectual"

ulala, @sharky6000.bsky.social

2.5 hours of building alpha for from scratch from a podcaster with indian name

what don't you know about the internet

to be a little reminiscent about the past, it's been 4 years since I tried to achieve something in the field I like all for nought, and I don't really know what's next, tbh

Transformers as Implicit State Estimators: In-Context Learning in Dynamical Systems Usman Akram, Haris Vikalo Action editor: Yoshinobu Kawahara https://openreview.net/forum?id=hIMK5MvGkP #kalman #transformers #prediction

📢 Excited to share that our paper 𝗧𝗿𝘂𝘀𝘁 𝗥𝗲𝗴𝗶𝗼𝗻 𝗜𝗻𝘃𝗲𝗿𝘀𝗲 𝗥𝗲𝗶𝗻𝗳𝗼𝗿𝗰𝗲𝗺𝗲𝗻𝘁 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 (𝗧𝗥𝗜𝗥𝗟) has been accepted at @icmlconf.bsky.social 2026! 🌐 Project page: lnkd.in/dqNi68Mk 📄 Paper: lnkd.in/dzbcCbjU #ICML2026 #ReinforcementLearning #InverseRL #ImitationLearning #Robotics

Ulala, I've read about the paper on twitter, an it's an amazing paper

roger-creus.github.io/agentick/blog/ very cool work

what if you were a timbarr for my timbers

i've got an idea: what if we get rid of conferences and their rankings and leave only topical venues and workshops where people can discuss and claim their ideas. Those can be reviewed with people, AI, whatever. The goal is to connect people, to brainstorm without an achieving scores.