Researchers released the Taste-Bench benchmark, which measures long-term decision "taste" by freezing agent runs at decision forks in their trajectories and asking models to choose between two paths; it contains 502 questions drawn from coding and AI research tasks such as SWE-bench Pro and METR's MALT.
Research published in Science by Google DeepMind shows its AlphaProof Nexus independently solved 9 of 353 open Erdős problems, two of which had stood for 56 years, proved 44 conjectures in the OEIS integer sequence database, solved an open algebraic geometry problem, and improved known bounds for min-max optimization.
A research team led by PhAI Labs together with Stanford, Oxford and Princeton has released JEPA-Anything, extending Yann LeCun's JEPA architecture into a general-purpose world model that works across seven domains.
A benchmark tested whether a local Qwen3.8-27B-Q4_K_M.gguf model can produce exact results of positive integer addition using only English words, with 5,070 no-reasoning cases and 169 reasoning-enabled control cases. With reasoning off, numerical accuracy was only 23.57% — 97.04% for 1-3 digit problems but 6.44% for 10-13 digits, with 96.17% format compliance; with reasoning on, it answered 167 of 169 attempts correctly.
A research team from Carnegie Mellon, MIT, NYU, and Stanford has developed an AI called Ataraxos that defeated Pim Niemeijer, considered the strongest Stratego player in history, with 15 wins, 1 loss, and 4 draws, using only 16 GPUs and a few thousand dollars for training. Stratego is an imperfect-information game with large amounts of hidden information and very long game timescales; even DeepMind had previously failed to build a machine that could reliably beat top human players.
Ataraxos, developed by a research team, defeated four-time world champion and widely acknowledged greatest Stratego player Niemeijer with an 85% effective win rate over three weeks of play, and won 38 of 40 games in an exhibition match at the 2025 world championship, using less than $8,000 of compute to crack imperfect-information games.
Why it matters: The article compares Ataraxos and DeepNash training costs and methods, giving readers insight into the latest progress in imperfect-information game research.
OpenAI shared an AI-generated solution approach to the Navier–Stokes Millennium Prize Problem, including a solution write-up and a Lean formalized proof. The material comes from an official OpenAI announcement; the feed does not include the full text, so specifics of the proof and methods should be checked via the original link.
Why it matters: OpenAI has publicly released an AI-generated approach to the Navier–Stokes Millennium Prize Problem with a Lean formalized proof, letting readers track AI's actual output in mathematical proof work.
Hugging Face proposes Quantization-Aware Healing (QAH), compressing GPT-OSS 120B to a 60B-parameter structure and quantizing it to MXFP4, then distilling directly from the original pre-compression model rather than from a recovered bfloat16 checkpoint.
DiG-bench (Discovery in Games) has been released, a benchmark of 70 games for testing AI's ability to autonomously discover hidden rules through exploration; 21 of the games are publicly available.
Google Research released the WikiProfile benchmark (2,150 Wikipedia facts, each with 10 tasks) and a knowledge profiling framework. After evaluating 13 LLMs, it found that frontier models have 95–98% of facts encoded but still fail to directly recall 26–34%, and 11–12% still fail even with thinking enabled.
Google Research published a reinforcement learning-based quantum error correction control framework in Nature, letting an agent continuously tune thousands of control parameters using error detection events to counter drift during computation. In experiments injecting artificial drift on the Willow superconducting processor, RL control improved logical stability by 3.5x and further reduced the logical error rate by 20% on top of expert calibration.
A Google Research paper at COLM 2026 found that enabling chain-of-thought reasoning allows Gemini-2.5 (Flash and Pro) and Qwen3-32B to recall simple single-hop facts they nearly always fail to answer when reasoning is turned off.
Berkeley AI Research proposes GRASP, a gradient-based world model planner that makes long-horizon planning more practical. Three core improvements: lifting trajectories into virtual states so optimization runs in parallel over time, injecting randomness directly into state iterations to support exploration, and reshaping gradients to avoid traversing fragile state-input gradients of high-dimensional vision models.
Google Research introduced TurboQuant, a compression algorithm (to be published at ICLR 2026) that combines Quantized Johnson-Lindenstrauss (QJL) and PolarQuant (to be published at AISTATS 2026) to achieve zero-memory-overhead vector compression, compressing models and KV cache without accuracy loss.
Google Research, with Cornell and Harvard, published a PNAS paper in which 12 high-temperature superconductivity experts wrote 67 questions and blind-reviewed answers from 6 LLMs.
Berkeley AI Research proposes an off-policy RL value-learning algorithm based on divide and conquer that no longer uses temporal difference (TD) learning, whose Bellman recursion error accumulation makes it hard to scale to long-horizon tasks.
Berkeley AI Research proposes a quantitative theory of the word2vec learning process: under several mild approximations, its learning problem reduces to unweighted least-squares matrix factorization, the gradient flow dynamics have a closed-form solution, and the final embeddings are equivalent to PCA on the target matrix M*.