Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs... Beyond the Base Model? New paper questions the common assumption that RLVR helps LLMs acquire novel reasoning abilities.show more

ICML 2026: Latent Reasoning in TRMs is Secretly a... Policy Improvement Operator Why does recursive reasoning, especially latent reasoning, actually work? The theory is still young, and even mechanistic explanations are limited. We close part of this gap by showing that latent reasoning is secretly doing policy improvement. Each recursion pushes the model steadily toward the target. Based on this view, we propose an algorithm that boosts learning and inference efficiency by up to 18x.show more

New paper! LLMs Corrupt Your Documents When You Delegate... LLMs are enabling a new way of working: delegated work, where users supervise an LLM as it edits documents on their behalf. Delegation requires trust: does the LLM complete tasks without introducing errors? We simulate delegation across 52 professional domains and find that LLMs Corrupt Your Documents When You Delegate. 🧵1/Nshow more

Current Vision-Language-Action (VLA) paradigms in autonomous driving primarily rely... on Imitation Learning (IL), which introduces inherent challenges such as distribution shift and causal confusion. Online Reinforcement Learning offers a promising pathway to address these issues through trial-and-error learning. However, applying online reinforcement learning to VLA models in autonomous driving is hindered by inefficient exploration in continuous action spaces. MindDrive, a VLA framework comprising a large language model (LLM) with two distinct sets of LoRA parameters. The one LLM serves as a Decision Expert for scenario reasoning and driving decision-making, while the other acts as an Action Expert that dynamically maps linguistic decisions into feasible trajectories. Paper Title: MindDrive: A Vision-Language-Action Model for Autonomous Driving via Project: Link:show more

Try 👁 Agentic Vision with Gemini 3 Flash in... Google AI Studio or Vertex AI. This new capability enables the model to effectively use code and reasoning to improve performance for common vision tasks. See Agentic Vision in action:show more

RL is back! But is it always the best... choice? In a new paper, we investigate under what circumstances neuroevolution outperforms RL in transfer learning tasks. See more details in the thread below 🧵 While NE performs best in simpler domains, it will be interesting to see if the lessons learned here can also be applied to more complex systems/tasks (LLMs?).show more

Reinforcement Learning (RL) has long been the dominant method... for fine-tuning, powering many state-of-the-art LLMs. Methods like PPO and GRPO explore in action space. But can we instead explore directly in parameter space? YES we can. We propose a scalable framework for full-parameter fine-tuning using Evolution Strategies (ES). By skipping gradients and optimizing directly in parameter space, ES achieves more accurate, efficient, and stable fine-tuning. Paper: Code:show more

Diffusion models are great, but we can squeeze out... so much more from them. The only problem is that it usually requires extra training or manual representation editing. In our new paper, we show that with the current capabilities of LLMs, it is much simpler than we thought!show more

Meet #DBRX: a general-purpose LLM that sets a new... standard for efficient open source models. Use the DBRX model in your RAG apps or use the DBRX design to build your own custom LLMs and improve the quality of your GenAI applications.show more

Contemporary discussion (hype?) about LLMs and “pausing AGI development”... seems oblivious of Moravec’s paradox. We’ve hypothesized since the 80s — that the hardest problems in AI involve sensorimotor control, not abstract thought or reasoning. It explains why AI is mastering games, dialogue, and scene generation. But robots are conspicuously missing from this revolution. Today, we announce 2 major advancements on the embodied road to general intelligence:show more

What if we could teach an AI to master... the strategic game of 2048 through pure reinforcement learning? I did exactly that with "Agent 2048" - fine-tuning Qwen 7B model using GRPO to develop spatial reasoning and merge strategies with zero prior gameplay SFTdata! Thanks to Hugging Face and Unsloth AI for their easy to use implementation kalomaze and will brown you might like this :)show more

A viral paper "Language Model Represents Space and Time"... recently claims that LLMs learn "world models". As much as I like Max Tegmark's works, I disagree with their definition of world model. World model is a core concept in AI agent and decision making. It is our mental simulation of how the world works given interventions (or lack thereof). A world model captures causality and intuitive physics, telling the agent what is likely and what is impossible. It can and should be used for counterfactual reasoning, i.e. "what ifs": what would happen if I knock over a cup of water? Where would I have been if I had not taken that bus? Yann LeCun Yann LeCun says it well in his position paper ( I quote: "Using such world models, animals can learn new skills with very few trials. They can predict the consequences of their actions, they can reason, plan, explore, and imagine new solutions to problems. Importantly, they can also avoid making dangerous mistakes when facing an unknown situation." The first use of the term World Model in deep policy learning is attributed to hardmaru & Jürgen Schmidhuber: In their seminal paper, an agent masters shooting skills in the popular game Doom (demo below) by learning in imagination, using an internal world model as a "physics simulator". To put in a simple Python math formula, world model learns a function F(s[0:t-1], a) -> s[t:], which takes as input the observed past and current action, and outputs plausible future states. Now the definition of World Model in Tegmark's paper seems to be about predicting GPS coordinates and time eras. I see this as just a classification task with no causal learning and simulation going on. You cannot make meaningful interventions against that model, nor can you optimize any decision making in a closed feedback loop. As for the "space & time neurons", I think they are most similar to the "sentiment neuron" that OpenAI published in 2017: Predicting GPS is conceptually no different from predicting sentiment in my opinion. I don't think their experimental results are wrong - just that their conclusion is on shaky grounds. I welcome any debate! Paper link:show more

GPT-5.5 by Reasoning Effort: I've asked it in Codex... to create a physics-based visualisation of RL cycles for different sized models (70b, 1t, 10t), to demonstrate how the amount of RL you can do differs by model size. My assessment of each: - Low: weird slop - Medium: kinda cooked - High: sort of tried but ultimately incoherent - Extra High: elite - really nice idea and well executed Obviously this is just one shot, but worth trying different reasoning levels for the new models, medium seems to be pretty good for GPT-5.5 and it was really bad for many previous GPT models.show more

Brilliant open-source tool from Anthropic Visualize internal LLM pathways... through open attribution graphs. → Attribution graphs reveal which neurons and features drive each output token, turning the LLM’s black box into a clear map. They open-sourced a library that generates attribution graphs to trace how LLMs arrive at outputs. These graphs reveal internal activation flows and feature influences in popular open-weights models. 🧠 Attribution Graphs Graphs link model activations to output tokens. Nodes represent neurons or features. Edges show influence strength. Researchers inspect paths to see which components drive each decision. 🛠️ Toolchain and Frontend Circuit-tracer library works with models like Gemma-2-2b and Llama-3.2-1b. Demo notebooks illustrate multi-step reasoning examples. Neuronpedia hosts an interactive UI for graph exploration, annotation, and sharing. 🔍 Hypothesis Testing Users tweak feature values in graphs, regenerate outputs, and observe changes. This lets researchers validate whether specific circuits control behaviors like language translation or reasoning chains.show more

🚀New paper out - We present Video-MSG (Multimodal Sketch... Guidance), a novel planning-based training-free guidance method for T2V models, improving control of spatial layout and object trajectories. 🔧 Key idea: • Generate a Video Sketch — a spatio-temporal plan with background, foreground, and motion in the pixel space. • Encode this structure directly into the latent space of the diffusion model during generation, which does not require fine-tuning or additional memory during inference. 🧵show more

BIOS Dev Update: Fast Chat, Thinking Traces & More... What's New: • Fast Chat Mode: A faster pipeline for simple questions that switches to literature search only when needed. It runs in an agentic loop, searching as many times as it needs until it has enough context to give you a complete answer. • Thinking Traces on Screen: BIOS now shows its reasoning on the main loading screen while Deep Research is running, so you can follow along in real time.show more

as an intro to mechanistic interpretability, i decided to... look into the formation of induction heads, which are circuits that allow LLMs to perform in-context learning by searching for previous occurrences of a sequence to predict the next token to form these circuits i trained attention-only transformers to repeat varying sequence lengths of random tokens. by randomizing the sequence length, i prevented the models from relying on rote memorization, forcing them to instead develop a generalizable circuit during this dive, i recorded some really cool findings and saw some interesting visual patterns emerging:show more

🚀 We introduce Neural Theorizer (NEO) — a new... type of world model that learns to theorize the world from observation, without language or LLM supervision. Selected as an ICML 2026 oral presentation — 0.7% of submitted papers. The paper asks: "What does it mean to understand the world and build a world model?" Today’s world models are often trained to predict the future: the next frame, next latent state, or next observation. But is prediction enough? We argue that a world model should be a theory-building system: one that discovers reusable primitives, composes them into executable explanations, and transfers those explanations to novel phenomena. NEO is our first step toward this vision — a World Theory Model that learns explicit, compositional theories from raw observation. This work was led by my wonderful students: Doojin Baek*(Doojin Baek), Gyubin Lee* (GyuBin Lee), Junyeob Baek (Junyeob Baek), and Hosung Lee (Hosung Lee). For more details, take a look at the paper — and if you’re attending ICML, let’s talk there! 📄 arXiv: 🌐 Project page:show more

GROK 4.20 TO TRANSFORM TRADING: THE ONLY AI THAT... BEATS THE MARKET Six AIs entered the stock trading arena; one is set to change the whole game. * Grok 4.20 finished +21%, crushing every other model in the competition * Outperformed GPT-5.1, Claude 4.5, Gemini 3 Pro, all ending in the red * Used a “monk mode” strategy: low risk, high reasoning, zero hype * Doubled the average annual return of an S&P 500 index fund in just 14 days * The results mean Grok isn’t just reacting to markets; it's on track to move them * Elon: “Grok 4.20 is the best quant trader we’ve built yet, profits while others fail” This isn’t paper trading. It’s real money, real markets, real dominance. Source: xAI, CNBCshow more