Google DeepMindFirst party
Investing in multi-agent AI safety research
Google DeepMind and partners announce a $10M funding call for multi-agent safety research.
AI news for builders and product teamsUpdated Oct 1, 2026, 10:00 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest firstImport AIResearch
Import AI 460: Reward hacking society, RSI data from Anthropic; and RL-based quadcopter racing
Researchers built SocioHack, a 72-environment benchmark showing RL-trained models rediscover historically patched regulatory loopholes; Anthropic reports an 8x increase in code merged in 2026 versus 2021-2024; and RL-trained quadrotors beat a champion human pilot.
LLM Research Papers: The 2026 List (January to May)
A curated roundup of notable LLM research papers that came out this year
Import AIResearch
Import AI 459: AI oversight is difficult; scaling laws for protein folding models; and pricing the extinction risk of AI systems
Import AI 459 covers a paper estimating the US AI economy's quality-adjusted output grew roughly 2,290 percent in 2024 and 2,271 percent in 2025, UK AISI research on why automated alignment is difficult, Stanford-led release of the 100M-image GPIC dataset, and Biohub's ESMFold2 protein model.
Import AIIndustry
Import AI 458: Reckoning with the future; and a singularity story
Import AI issue 458 consists of a long essay based on a 2026 Cosmos HAI Lab Lecture given at the University of Oxford, arguing that continued AI progress forces a choice between exploring the future or retreating from the present, plus a fictional story about a positive singularity.
Import AIResearch
Import AI 457: AI stuxnet; cursed Muon optimizer; and positive alignment
Import AI 457 covers SentinelOne's teardown of the fast16.sys virus that tampered with high-precision calculation software, Tilde Research's finding that the Muon optimizer can cause neuron death in MLP layers and its proposed Aurora optimizer, a position paper on positive alignment, and Prime Intellect tests of autonomous AI research agents on the nanoGPT speedrun.
Recent Developments in LLM Architectures: KV Sharing, mHC, and Compressed Attention
From Gemma 4 to DeepSeek V4, How New Open-Weight LLMs Are Reducing Long-Context Costs
Import AIResearch
Import AI 456: RSI and economic growth; radical optionality for AI regulation; and a neural computer
Researchers with the Institute for Law & AI propose "radical optionality" for AI governance, urging governments to build institutions and legal authorities now while avoiding overregulation. Separately, a Meta and KAIST paper explores neural computers, and economists model how automating AI research could produce explosive economic growth.
Berkeley AI ResearchFirst party
Adaptive Parallel Reasoning: The Next Paradigm in Efficient Inference Scaling
Overview of adaptive parallel reasoning. What if a reasoning model could decide for itself when to decompose and parallelize independent subtasks, how many concurrent threads to spawn, and how to coordinate them based on the problem at hand? We provide a detailed analysis of recent progress in the field of parallel reasoning, especially Adaptive Parallel Reasoning. Disclosure: this post is part landscape survey, part perspective on adaptive parallel reasoning. One of the authors (Tony Lian) co-led ThreadWeaver ( Li
Import AIResearch
Import AI 455: AI systems are about to start building themselves.
Import AI editor Jack Clark writes that there is a 60%+ chance that no-human-involved AI R&D, where a system could autonomously build its own successor, happens by the end of 2028, citing benchmark trends in coding, reproducibility, ML engineering, and kernel design.
Berkeley AI ResearchFirst party
Gradient-based Planning for World Models at Longer Horizons
GRASP is a new gradient-based planner for learned dynamics (a “world model”) that makes long-horizon planning practical by (1) lifting the trajectory into virtual states so optimization is parallel across time, (2) adding stochasticity directly to the state iterates for exploration, and (3) reshaping gradients so actions get clean signals while we avoid brittle “state-input” gradients through high-dimensional vision models. Large, learned world models are becoming increasingly capable. They can predict long sequenc
Berkeley AI ResearchFirst party
Identifying Interactions at Scale for LLMs
--> Understanding the behavior of complex machine learning systems, particularly Large Language Models (LLMs), is a critical challenge in modern artificial intelligence. Interpretability research aims to make the decision-making process more transparent to model builders and impacted humans, a step toward safer and more trustworthy AI. To gain a comprehensive understanding, we can analyze these systems through different lenses: feature attribution , which isolates the specific input features driving a prediction (
Berkeley AI ResearchFirst party
Information-Driven Design of Imaging Systems
An encoder (optical system) maps objects to noiseless images, which noise corrupts into measurements. Our information estimator uses only these noisy measurements and a noise model to quantify how well measurements distinguish objects. Many imaging systems produce measurements that humans never see or cannot interpret directly. Your smartphone processes raw sensor data through algorithms before producing the final photo. MRI scanners collect frequency-space measurements that require reconstruction before doctors ca
Berkeley AI ResearchFirst party
RL without TD learning
In this post, I’ll introduce a reinforcement learning (RL) algorithm based on an “alternative” paradigm: divide and conquer . Unlike traditional methods, this algorithm is not based on temporal difference (TD) learning (which has scalability challenges ), and scales well to long-horizon tasks. We can do Reinforcement Learning (RL) based on divide and conquer, instead of temporal difference (TD) learning. Problem setting: off-policy RL Our problem setting is off-policy RL . Let’s briefly review what this means. Ther
Berkeley AI ResearchFirst party
What exactly does word2vec learn?
What exactly does word2vec learn, and how? Answering this question amounts to understanding representation learning in a minimal yet interesting language modeling task. Despite the fact that word2vec is a well-known precursor to modern language models, for many years, researchers lacked a quantitative and predictive theory describing its learning process. In our new paper , we finally provide such a theory. We prove that there are realistic, practical regimes in which the learning problem reduces to unweighted leas