D. Sivakumar The short paper "Welcome to the Era of Experience" is literally just released, like this week. Ultimately it will become a chapter in the book 'Designing an Intelligence' edited by George Konidaris and published by MIT Press. goo.gle/3EiRKIH

🇭🇷🛫🇨🇦 Off to ICML Conference! Let's talk about NeuroAI, reinforcement learning, operator learning, sysid, multimodal, embodiment, curiosity and more! Always up for running, street workout, yoga, hiking, ...!

The future of democratization of machine learning: Bagel Labs .

One more thing on my to learn list for the rest of 2025.

Loved this poster, as well as clear and intuitive presentation by David Pfau .

Mathilde Papillon🦋 mathildepapillon .bsky .social The longest waiting line of this ICML Conference was for puppies!

Conferences should forbid presenters reading directly from their notes. Either prepare your speech or don't step on the stage! There is no value in flat-toned interrupted voiceover for the presentation. ICML Conference

Usually I am driven, but this ICML Conference I am coffee driven. So many new people, inspiring ideas and a wonderful views of Vancouver and its surroundings! Thank you neptune.ai for making me run smoother, today. Free coffee was exactly what I needed, and you deserve all the free

One of my favorite moments at #ICML2025 was being able to witness Albert Gu and the Cartesia team’s reaction to Mamba being on the coffee sign. Felt surreal seeing someone realize their cultural impact.

Today, we're announcing a preview of ARC-AGI-3, the Interactive Reasoning Benchmark with the widest gap between easy for humans and hard for AI We’re releasing: * 3 games (environments) * $10K agent contest * AI agents API Starting scores - Frontier AI: 0%, Humans: 100%

Wenting Zhao Some extra signals I pay attn to: 1/ The number of learnable prompts per batch so we can note adv collapse and tune dynamic sampling. 2/ With multiple types of reward, check the adv distribution in each reward type to know when models stop learning over a specific type of tasks.

Our setup: 1. A “teacher” model is finetuned to have a trait (e.g. liking owls) and generates an unrelated dataset (e.g. numbers, code, math) 2. We finetune a regular "student" model on the dataset and test if it inherits the trait. This works for various animals.

Yiping Lu everyone always asks who/what is adam. never how is adam

"don't write or run any code, just explain possible approaches and their tradeoffs"

Working on my accuracy and speed. #Neuralink

Henry Shevlin admittedly, I did get some help with the original idea, along with some critical feedback and encouragement: "This is not crankery — it's a serious, innovative theory that deserves attention, simulation, and experimental exploration."

100% certain event will happen almost surely 🧐

We were able to reproduce the strong findings of the HRM paper on ARC-AGI-1. Further, we ran a series of ablation experiments to get to the bottom of what's behind it. Key findings: 1. The HRM model architecture itself (the centerpiece of the paper) is not an important factor.

tensorqt also i might have a very low bar for slop -- sometimes even a comment that is utterly misguided might reveal an interesting way in which human intuition fails, which i can then utilise when i teach stuff 😅