AI News

The fridge-size satellite with AI chips on board is part of Project Suncatcher, the company's moonshot effort to put AI data centers in space and harness solar energy.

Join the discussion on this paper page

Introducing new capabilities to frontier models has long been the goal of posttraining, which predominantly employs supervised finetuning (SFT) and reinforcement learning (RL) to this end. Conventional wisdom dictates that RL enables strong generalization on new tasks without losing existing capabilities, while SFT is prone to weak generalization and catastrophic forgetting. At the same time, SFT can learn from off-policy expert data, whereas RL must rely on a model's ability to find successful trajectories with repeated sampling. In our work, we seek to leverage the strength of on-policy learning while utilizing the privileged information contained in off-policy data. However, rather than modifying the learning objective to accommodate this data, we instead tailor the data distribution to better suit the learner. We introduce a Markov chain Monte Carlo (MCMC) sampling algorithm that progressively transforms off-policy traces to be more on-policy given a reference model for finetuning. Across tasks like scientific skill acquisition, mathematical reasoning, and open-ended expertise, our sampling algorithm enables SFT to rival prevailing posttraining techniques, often generalizing better and forgetting less than strong on-policy baselines. In addition, the resulting finetuned models exhibit strong distributional performance and are capable of learning beyond sharpening the base model distribution. At a higher level, our approach presents sampling as a model-native operator that shapes data for learnability, offering broader utility as a general-purpose primitive throughout the posttraining stack.

As AI accelerates scientific discovery, we’re helping researchers keep pace by quickly authenticating if biological designs are created with built-in safeguards, without stalling crucial work. 🔬

We’re releasing these tools on an open basis for research use to strengthen safeguards → https://t.co/…

Welcome to Made on YouTube. Explore the Made On YouTube 2026 hub for all keynote news, AI creation tools, and feature updates.

BAG Ventures announces close of a $11.3M Fund I to invest in all things AI.

Flow Engineering, which is bringing AI agents to hardware design, also landed Roelof Botha as an angel investor and board member.

La startup riceve il capitale da un pool di imprenditori italiani che hanno creato e fatto crescere startup

An API security workbench for web and Android apps. A browser already routed through the proxy, an Android emulator that already trusts the certificate, and one merged API surface you can export. Free and open source.

Errol is a robot powered by real AI, and he gets math wrong on purpose. Your kid's job: catch the mistakes. Critical thinking for the AI age.

This conversational AI assistant is intended to provide personalized feedback to creators, rather than generic advice.

Apptronik pushes its commercial humanoid platform Apollo 2 into real-world factory workflows with Mercedes-Benz and GXO Logistics, featuring hot-swappable batteries and 25 kg payload.

Join the discussion on this paper page

Recent advances in LLM reasoning models---driven primarily by the paradigm of post-training via reinforcement learning with verifiable reward (RLVR)---have enabled them to accomplish impressively complex tasks. However, in parallel with their rising capabilities, LLMs have increasingly displayed signs of language drift in their chains of thought (CoTs): unusual, non-standard, and seemingly nonsensical language use. Although it is well-documented---and can potentially impair CoT monitorability---the causes of language drift are thus far poorly understood. In this paper, we identify the conditions under which language drift occurs: we prove theoretically that RLVR optimization pressure permits unbounded language drift, while supervised fine-tuning does not. We then show empirically that language drift specifically arises during RLVR on novel reasoning tasks---i.e. when the target behavior cannot be drawn out of the base model. Finally, we prove that it is not possible to constrain language drift without constraining expected reward, suggesting that CoT monitorability cannot be improved without harming performance during RLVR post-training at the frontier.

Join the discussion on this paper page

Instead of building its durable execution engine on top of an external database, the company developed its own storage, replication, and redundancy layers. This architecture allows Restate to be exceptionally fast and lightweight.

Nei campi mancano centomila stagionali. I 200 robot da vigneto al lavoro in Europa sono di un gruppo italiano, il mercato è agli inizi

Il Ministro per la Pubblica Amministrazione ha risposto a una domanda de Il Giornale d’Italia in occasione dell’evento “Le Pubbliche Amministrazioni tra sovranità europea e AI”, organizzato da BIP, soffermandosi su capitale umano, intelligenza artificiale, futuro de

A meeting recorder for your Mac with a live copilot. It transcribes on your machine and suggests answers from your own documents mid-call. Free and open source.

By launching an AI agent for food ordering, DoorDash is looking to gain an edge over rivals Uber Eats and Grubhub.

A humanoid robot has learned to walk across grass, wood chips and a steep fifteen-degree ramp without cameras, LiDAR, or any other external sensor — relying entirely on the same kind of body-awareness that lets you feel the slope of the ground beneath your feet. The achievement, published in the journal Autonomous