Beyond De-Skilling: Intelligence Explosion and the End of Skill as a Stable Category 4m4.it/posts/beyond-de-skill…

Brandt, S. (2026). Theory of mind and language development. Cambridge University Press. doi.org/10.1017/978100949629…

Can a vision model learn to see with no augmentations, no masking, no cropping, no reconstruction? 🎬 It can! Introducing Temporal Difference in Vision (TDV), a new visual representation learning paradigm built on a single assumption: the past causes the future. TL;DR : - We introduce TDV, the first approach to learn useful representations without any augmentations, masking, cropping or pixel based reconstruction. - TDV matches SOTA recipes like DINO and iBOT on dense spatial tasks - We also show that as data scales up, weaker assumptions work better. 🧵Thread:

[LG] How Post-Training Shapes Biological Reasoning Models L Fesser, H Zhang, M M. Li, E Wang… [Harvard University] (2026) arxiv.org/abs/2606.16517

Neural Variability Enhances Artificial Network Robustness Robin Preble, Praveen Venkatesh, Stefan Mihalas, Kameron Decker Harris arxiv.org/abs/2606.13801 [𝚌𝚜.𝙻𝙶 𝚚-𝚋𝚒𝚘.𝙽𝙲]

[LG] ExpRL: Exploratory RL for LLM Mid-Training V Xiang, A Setlur, C Blagden, N Haber, A Kumar [Stanford University & CMU & OpenAI] (2026) arxiv.org/abs/2606.17024

Gefen: Optimized Stochastic Optimizer Nadav Benedek, Tomer Koren, Ohad Fried arxiv.org/abs/2606.13894 [𝚌𝚜.𝙻𝙶 𝚌𝚜.𝙰𝙸 𝚌𝚜.𝙲𝙻 𝚌𝚜.𝙲𝚅]

Agent trajectories as programs: fingerprinting and programming coding-agent behavior Hamidah Oderinwale arxiv.org/abs/2606.16988 [𝚌𝚜.𝚂𝙴 𝚌𝚜.𝙻𝙶]

Specifications for Humans, Agents, and Tooling Mark Marron arxiv.org/abs/2606.15084 [𝚌𝚜.𝚂𝙴]

Natively Unlearnable Large Language Models Gaurav R. Ghosal, Pratyush Maini, Aditi Raghunathan arxiv.org/abs/2606.13873 [𝚌𝚜.𝙻𝙶 𝚌𝚜.𝙲𝙻]

Next-token prediction is myopic. What if transformers learn to predict their own next latent state? 🌠 We present 𝗡𝗲𝘅𝘁-𝗟𝗮𝘁𝗲𝗻𝘁 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻 (𝗡𝗲𝘅𝘁𝗟𝗮𝘁): a self-supervised learning method that teaches transformers to form compact world models for reasoning and planning. It also unlocks up to 3.3x faster inference via self-speculative decoding! 🚀

🚨 New Preprint! 🧠 We gave an AI model one simple rule: rearrange your neurons so that nearby ones respond alike. We never told it what a face, a voice, or a sentence was. It grew brain-like maps for all three anyway. 🧵👇 🌐 Website: topo-omni.epfl.ch

The New Cognitive Infrastructure of Science – Random Bits of Knowledge 4m4.it/posts/new-cognitive-i…

Modern text-to-image models are increasingly powered by large pretrained LLMs. But there is a curious mismatch: the LLM typically encodes the prompt only once, while the evolving noisy latent states are handled entirely by a newly trained generative backbone. Can pretrained multimodal prior participate in the denoising process? Introducing RepFusion. (1/12) 📄 arxiv.org/abs/2606.14700 🌐 xichenpan.com/repfusion/

"Learning the Geometry of Data: A Mathematical Review of Shape Space Analysis" (by Gary P. T. Choi, Khanh Dao Duc, Shira Faigenbaum-Golovin, Karen Habermann, Emmanuel Hartman, Christoph von Tycowicz, Chi Zhang, Wenjun Zhao, Felix Zhou): arxiv.org/abs/2606.17022

VLMs Systematically Fake Visual Understanding Even when VLMs appear to be good at visual understanding, most of their answers are not actually grounded in the image (hallucinated!). We identify two types of hallucinations that appear in up to 98% of answers that seem to demonstrate visual understanding. First, textual biases. The model answers using language patterns, information in the question, and knowledge learned during training, without engaging its visual representations. Second, spurious images. The model constructs false visual content inside its internal representation and then answers as if this imagined content were grounded in the real image. In both cases, the answers may still be correct, but they are not grounded in the visual input at all!!

Cédric Villani le mois dernier : les LLM ne sont pas intelligents et ne comprennent rien à ce qu'ils racontent, ce ne sont que des machines statistiques réductibles à des fonctions ; la preuve de leur inintelligence, ces modèles se trompent sur l'exemple de la voiture à laver...

1/ Standard transformers have a fundamental topological flaw: they cannot track dynamic states over time without running out of layers. Once a state representation reaches the top layer of the feedforward stack, the model's ability to update its belief collapses. 🧵

The alpha version of my new book "Optimal Transport for Machine Learners" is out, with in particular an online version with interactive figures gpeyre.com/ot4ml/

When a Nobel Laureate Uses an LLM to Prove a Theorem: A Turning Point for Mathematical Discovery – Random Bits of Knowledge 4m4.it/posts/when-a-nobel-la…

🚨🌍World models are surprisingly fragile! We introduce BadWorld, an adversarial attack for visual world models. A tiny perturbation to the starting image 🖼️ can break down the whole world. Code:github.com/LinghuiiShen/BadW… Paper:huggingface.co/papers/2606.1… Arxiv:arxiv.org/abs/2606.16519