Research papers, technical reports, benchmarks and datasets.
In Nature this week, Satya Nadella says his organization’s new research shows how machine learning can improve weather forecasts two to six weeks out, a timeframe that has historically been very difficult to predict. He highlights the work’s potential to help people better prepare for extreme weather.
NVIDIA and AI4Bharat at IIT Madras introduce Nemotron-pretraining-Indic-v1, approximately 2.8 trillion tokens across 22 Indian languages, releasing soon alongside its open, GPU-accelerated generation pipeline. The corpus includes native-script text for all 22 languages and romanized text for twelve, translated from curated English sources with quality checks and source tracing. An Indic continued-pretraining stage improved five reported benchmarks, but combined synthetic and natural data, so the comparison does not isolate this corpus’s contribution.
Why it matters: The workflow offers a reusable approach to multilingual data generation by separating translation, quality assessment, repair, and script conversion while preserving links to source documents.
10 new papers in health science research. all written with muse models in under 4 weeks. health is one of the areas where we believe muse is going to have the greatest positive impact on humanity ❤️ https://t.co/eejFDFd2Co
Researchers used AI-based protein design tools to retain the catalytic cores of two types of protease while replacing the rest of each protein. The resulting enzymes could be tuned to cut specific proteins in half, and the researchers demonstrated their effectiveness by cutting proteins associated with neurological diseases.
Microsoft researchers and collaborators published research in Nature on Probabilistic Bias Correction (PBC), an open-source machine learning tool that corrects systematic bias in existing weather models to improve forecasts two to six weeks ahead. In the study, PBC improved ECMWF’s physics-based model accuracy for 98 percent of precipitation and 92 percent of temperature targets, and doubled the accuracy of its AI forecasting system. The tool can run on CPUs, though ECMWF cautioned that operational implementation remains challenging.
Researchers from Sakana AI and UC Berkeley introduced MASS, a method that uses copies of Qwen3.6-27B to solve tasks, evaluate results, and optimize workflows without a stronger teacher model. After two rounds, its test-task win rate against the base model reached 69.9 percent, although benchmark gains largely reflected more completed work and some scores declined. Shared blind spots allowed flawed results to pass evaluation and enter training, prompting the researchers to recommend independent checks, limits on autonomy, and clear intervention conditions.
Microsoft Research introduces StudentSim, a research system that trains personalized student simulators from publicly available, de-identified learner records to evaluate and improve AI tutors. StudentSimEval measures behavioral fidelity and guidance responsiveness across chess, second-language English writing, and mathematics, where StudentSim outperforms leading baselines. In a chess tutor study, training with StudentSim feedback achieves 90.5% accuracy versus 75.7% without reinforcement learning and 71.6% with GPT-5.4 feedback; real learner studies remain essential, and the paper and code are publicly available.
Why it matters: Evaluating both a simulated learner’s starting ability and response to guidance offers a concrete method for assessing tutor feedback beyond the correctness of an explanation.
On October 7, OpenAI released 722 mathematics manuscripts, including two on the Kakeya conjecture: a 97-page paper claiming a three-dimensional maximal-function result stronger than Wang and Zahl's, and a 175-page paper claiming the four-dimensional Hausdorff dimension case. Both are attributed to OpenAI, are not Lean-formalized or peer-reviewed, and OpenAI's README says some unformalized results may be flawed.
Why it matters: The piece traces how two OpenAI manuscripts extend Wang and Zahl's three-dimensional Kakeya work to a stronger maximal-function version and to four dimensions, while noting the proofs lack peer review.
Columbia University's Li Yunzhu team presented progress on FlexiTac, an open-source flexible piezoresistive tactile sensor platform, at an IROS 2026 workshop. The sensor assembles in 3 minutes, ships with a full BOM and readout software, and integrates with the LeRobot robot-learning framework. Analog Devices is developing a commercial version with tactile resolution five times that of a human finger, produced via semiconductor manufacturing.
The ByteDance Seed team found that the DeepSeek-V4 series shows periodic fluctuations in long-context retrieval depending on where information appears, with a cycle of every 4 tokens. In the 128K needle-in-a-haystack test, moving the same information to a different position changed accuracy by as much as 40.2 percentage points.
IROS 2026 accepted 1,933 papers. A multi-label analysis counted about 809 papers in Robot Learning / Embodied AI and 564 in Navigation / Planning.
The NVIDIA team started with Nemotron 3 and used supervised fine-tuning, reinforcement learning, and a generate-verify-improve reasoning process to build specialized models that reached gold-medal thresholds at both IOI 2026 and IMO 2026.
Why it matters: The source gives a complete recipe for fine-tuning Nemotron 3 to gold-medal-level results at IOI and IMO, which readers can use to reproduce its data and reasoning workflow.
OpenAI published mathematical results from an unreleased internal frontier model through a blog, GitHub repository, and posts. The release includes 722 manuscripts grouped into 372 related result families, drawn from an evaluation of about 4,000 research questions. Each result used an average of about three hours of ChatGPT Pro reasoning compute.
Why it matters: OpenAI is producing mathematical results in bulk with an internal model, giving readers a view of the scale, compute cost, and outside criticism surrounding the work.
Google Research proposed using Google Earth AI's Population Dynamics Foundation Model (PDFM) as a plug-and-play geospatial foundation model for public health, enhancing existing epidemiological models without task-specific fine-tuning.
Researchers at PhAI Labs, Stanford, Oxford, and Princeton proposed JEPA-Anything. It splits JEPA's predicted state into four orthogonal factors, assigns a prediction module to each, and recombines them. It outperformed standard JEPA with the same architecture and data on ten test tasks in physics, robotics, weather, and other areas.
A benchmark tested the local Qwen3.8-27B-Q4_K_M.gguf on 5,070 cases without reasoning, asking it to add positive integers using only English words and return exact results. It achieved 23.57% numerical accuracy and 96.17% format compliance. Accuracy fell as the numbers grew, from 97.04% on one- to three-digit numbers to 6.44% on ten- to thirteen-digit numbers. In 169 paired cases with medium reasoning enabled, the model answered 167 correctly.
Google AI highlights work by AI research lab Living Models, which paired Gemma 4 with BOTANIC-1, a model trained on plant DNA from 320 species, to identify mutations that alter crop traits. In a melon yield gene test, the pair ranked the target mutation first out of 2,494 possibilities in under four minutes, versus years of physical breeding trials. A demo and paper are linked.
Hinton, Bengio, Andrew Barto, Jakub Pachocki, and 18 other authors published a paper titled 'What if automating AI R&D triggers an intelligence explosion?'
Meta said that after AI models achieved gold-medal-level performance in five math, physics, and chemistry competitions, it had mathematicians work with Muse Spark 1.1 and 1.2 in Thinking Mode through the meta.ai chat interface to solve open problems with no existing solution path and no custom research scaffolding.
Why it matters: Meta made six papers from mathematicians' collaboration with Muse Spark public and described the division of work between people and AI and the review process.
The Hugging Face team published a multi-harness RL guide, showing that the same model with identical weights can score 62% with one agent harness and 33% with another.
Why it matters: Scores for the same model differ by nearly half across agent harnesses. This open guide gives a multi-harness RL training path without changing the harness, along with measured comparisons.
ServiceNow CoreAI built AutoSynthData to find capability gaps from a target model's failures and a stronger teacher's successful trajectories, then generate and validate new tasks for post-training. As the model improves, the curriculum shifts to tasks that remain difficult. The process was demonstrated on the EnterpriseOps Gym dataset. Generated tasks must meet feasibility, realism, and difficulty requirements; validators must provide consistency, reliability, and completeness.
Ataraxos, an AI system developed by researchers at Carnegie Mellon, MIT, NYU, and Stanford, beat Pim Niemeijer, considered the strongest Stratego player ever, with 15 wins, one loss, and four draws. Training used only 16 GPUs and cost a few thousand dollars.
Microsoft Research developed an end-to-end machine-learning pipeline that uses L1 solar-wind observations and AE and Dst indices to generate localized space-weather risk estimates 30 to 60 minutes ahead for 66,935 substations in the contiguous United States.
A Google DeepMind team published a paper proposing SynthID Bio, a protein-watermarking system based on SynthID. It can distribute watermarks randomly through AI-designed protein sequences without harming protein function.
At an ECCV 2026 workshop, NVIDIA Research scientist Li Boyi presented two projects, MultiGen and DexImit. MultiGen uses diffusion models to add synchronized audio to videos generated by physical simulators.
DeepSeek published an exclusive technical article on Zhihu, explaining for the first time the DSec sandbox infrastructure behind all DeepSeek-V4 training, evaluation, and data preprocessing. The technical report is available on arXiv, jointly authored by DeepSeek and Tsinghua University with a team of more than 130 authors.
Anthropic says Claude discovered a previously unknown enzyme system in bacteriophage DNA, next to a stretch of repetitive DNA resembling CRISPR. The system's function is not yet known, but the few known systems with similar features can cut, copy, and paste DNA. Anthropic says more research is needed to determine what it does and whether it could be used for gene therapies like CRISPR.
Microsoft has published RetroChimera, a retrosynthesis-prediction framework, in Nature and open-sourced its implementation and weights. It combines the Transformer model R-SMILES 2 with the GNN model NeuralLoc, using a learned ensemble to rerank their predictions and performing strongly on both common and rare reaction types. In blind tests, chemists preferred its single-step reaction predictions. It found complete routes for nine of ten challenging targets, compared with five for a de novo model, four for an editing model, and two for NeuralSym.
Google Research has released MilleMiglia, a C++ instance generator that creates realistic, privacy-preserving benchmark data for middle-mile logistics and delivery problems. Its source code and documentation are available on GitHub.
Google Research proposes Retrieve-for-Train, a framework that uses offline RL to discover reward-aligned query fan-outs and compile them into supervision, then distills them into a 53.9M-parameter diffusion retriever. At inference, it generates a complete result set in one non-autoregressive pass without chain-of-thought tokens. The method fine-tunes fan-out language models based on Gemma3-4B and Qwen3-4B, using set-level attribute rewards to address paraphrase collapse in zero-shot LLMs and autoregressive latency.
Google Research proposes ToolGrad, a “answer-first, question-second” approach to generating tool-use data: first create a ground-truth tool-call chain, then work backward to infer the user's prompt, using just one LLM call.
Google DeepMind has released AlphaGenome Atlas, a platform containing predicted effects for 9 billion single-nucleotide variants in the human genome. At 1 PB, it is more than 30 times the size of the AlphaFold Database. From today, academic researchers can access it free through the web portal, AlphaGenome API, and a skill in Google Antigravity.
Why it matters: The article gives the scale of predicted effects for 900 million single-nucleotide variants and free academic access points, helping readers assess the availability of genome-interpretation resources.
In a paper published in Cell, Google, HHMI Janelia, Cambridge, and collaborators have published a connectome of the complete male fruit-fly brain and central nervous system. It contains more than 166,000 neurons and 125 million synaptic connections, making it the largest brain map to date by neuron count.
Why it matters: Google, HHMI Janelia, and partners have released the largest brain map by neuron count, showing a concrete application of AI in connectomics.
Hugging Face has released BenchMIRT, a multidimensional IRT method for auditing LLM benchmarks at the individual-prompt level and separating mixed ability signals within a benchmark. Trained on results from 100 LLMs across 16 benchmarks and more than 34,000 questions, it independently recovered two dominant dimensions—safety and general reasoning—without being told which benchmarks measured them. The analysis finds that scores on safety benchmarks such as BBQ and WMDP correlate more strongly with general reasoning, suggesting a single benchmark score may combine several signals.
Google Research has introduced MAPL-EMIT, a Swin-S transformer-based deep-learning framework that automatically detects, predicts, and localizes methane plumes from NASA EMIT hyperspectral satellite data. It achieved 84% recall on expert-labeled plumes.
Cleaning data and aligning the reward function for RLVR takes expertise and effort upfront, but the result is a model that's state-of-the-art on a complex task. Guest post by researchers at UIUC and Bridgewater, in collaboration with our team. https://thinkingmachines.ai/news/putting-task-expertise-into-rl
Google Research has introduced the experimental Planetary Prediction Engine (PPE) under Google Earth AI. From a natural-language query, it can autonomously discover geospatial data, engineer features, train and evaluate models, and produce a report—compressing weeks of manual data engineering into minutes.
Why it matters: The article reports quantified improvements for PPE in public health, food security, and epidemic forecasting, giving a view of the practical limits of geospatial modeling automation.
Google has released GlucoFM, a self-supervised foundation model with a dual-stream design that separates slow glucose trends from short-term fluctuations while preserving timing and missingness information. Across 14 evaluations spanning four cohorts and seven clinical tasks, it improved average PR-AUC by 5.8 percentage points over the best GluFormer variant and achieved the lowest MAE for predicting post-meal glucose response. It was pretrained on 109,066 hours of unlabeled CGM data and performed best at cross-dataset transfer and few-shot adaptation.
Multiverse Computing has published a paper proposing Quantization-Aware Healing (QAH). After compressing GPT-OSS 120B to 60B parameters and quantizing it to MXFP4, the method distills directly from the original model before compression rather than from a recovered bfloat16 checkpoint.
Why it matters: The paper presents a repair recipe that distills from the original model after compression and quantization, with stability comparisons against QAT.
METR research finds that AI's acceleration of science is uneven: cybersecurity-vulnerability reports sped up considerably from 2025 to 2026, mathematical research accelerated only slightly, and AI research itself showed no measurable acceleration. A university collaboration proposed SPADE, a self-play framework in which LLMs alternately generate executable training environments and solve them. Qwen3-30B-A3B scored 58.3 across the game-environment suite, an 8.1-point gain over its base model.
