AI Research | Tech AI Latest https://techailatest.news/ai-research Research that changes what AI can do: new training and inference methods, evaluation results, interpretability and safety work, and AI applied to science. Stories summarise the paper or lab report in plain language, name the authors and institution, and link to the original publication. en Wed, 28 Oct 2026 15:33:48 GMT https://techailatest.news/icon-512.pngAI Research | Tech AI Latesthttps://techailatest.news/ai-research Ai2 study finds LLM novelty judges flip verdicts when prompts change, from 93.5% to 40.9%https://techailatest.news/ai-research/ai2-llm-novelty-judges-prompt-sensitivity-studyhttps://techailatest.news/ai-research/ai2-llm-novelty-judges-prompt-sensitivity-studyTue, 27 Oct 2026 09:00:00 GMTTech AI LatestAI ResearchAi2 reported on 27 October 2026 that LLMs judging research novelty are prompt-sensitive: GPT-5.4 fell from 93.5% to 40.9% accuracy after one wording change.The Allen Institute for AI (Ai2) published a study on 27 October 2026 testing whether large language models can judge if a research idea is novel. Using ideas drawn from ICLR 2026 reviews, it found verdicts swing with small prompt changes: GPT-5.4's accuracy on human-versus-generated pairs fell from 93.5% to 40.9%. A simple Claude Opus 4.6 prompt beat purpose-built novelty checkers at a fraction of the cost. Data and code are open.

Read the full story

]]> Anthropic reports early drug development results with Claude across six partner organizationshttps://techailatest.news/ai-research/anthropic-claude-drug-development-six-partnershttps://techailatest.news/ai-research/anthropic-claude-drug-development-six-partnersTue, 27 Oct 2026 09:00:00 GMTTech AI LatestAI ResearchAnthropic published preclinical drug development results with Claude on 27 October 2026 from six partners including Bristol Myers Squibb and Enveda.Anthropic published on 27 October 2026 early results from drug development work with six partner organizations and its own program. Reported outcomes include a CAR T genetic component about 70% shorter than the standard design, roughly 35,000 antibody designs generated in about a month, antibody variants that bound more than 10 times more tightly, and an 89% cut in human review time for regulatory documents. Anthropic says the work is preclinical and that it plans to expand it.

Read the full story

]]> Google DeepMind maps the world's forest types at 10 m, correct 90% of the timehttps://techailatest.news/ai-research/google-deepmind-forty-global-forest-type-maphttps://techailatest.news/ai-research/google-deepmind-forty-global-forest-type-mapThu, 22 Oct 2026 14:30:00 GMTTech AI LatestAI ResearchGoogle DeepMind and partners released ForTy on 22 October 2026, a 10-metre global map of six forest and land-use types, correct 90% of the time in checks.Google DeepMind and Google Research, with the World Resources Institute and IIASA, released ForTy on 22 October 2026: a global map that sorts every 10-by-10-metre patch of land into six forest and land-use types for 2020, plus an early yearly series from 2017 to 2025. Checked at more than 9,000 random sites, it named the correct forest type 90% of the time. The map is free on Google Earth Engine and published in Scientific Data.

Read the full story

]]> Microsoft Research mathematicians prove the 31-year-old KLS conjecture with AI agent helphttps://techailatest.news/ai-research/microsoft-research-kls-conjecture-proof-ai-agentshttps://techailatest.news/ai-research/microsoft-research-kls-conjecture-proof-ai-agentsWed, 21 Oct 2026 09:00:00 GMTTech AI LatestAI ResearchMicrosoft researchers said on 21 October 2026 that they proved the 1995 KLS conjecture in convex geometry, using AI agents to explore about 100 research paths.Two Microsoft researchers, Xinzhi Zhang and Zhao Song, say they have proved the Kannan, Lovász and Simonovits (KLS) conjecture from 1995, showing a bound that does not depend on dimension. Microsoft Research described the result on 21 October 2026. The authors used parallel AI agents to explore about 100 research paths, while choosing the direction and verifying the proof themselves. The preprints are on arXiv and have not been peer reviewed.

Read the full story

]]> MIT-led team doubles accuracy of two-to-six-week weather forecasts with machine learninghttps://techailatest.news/ai-research/mit-machine-learning-subseasonal-weather-forecasts-naturehttps://techailatest.news/ai-research/mit-machine-learning-subseasonal-weather-forecasts-natureWed, 14 Oct 2026 15:00:00 GMTTech AI LatestAI ResearchAn MIT-led team reported in Nature that a machine learning correction doubled the accuracy of ECMWF forecasts two to six weeks ahead, beating long-term norms.Researchers from MIT, Harvard, the University of Toronto, ECMWF, Microsoft and others reported in Nature a machine learning method that corrects the errors of weather models two to six weeks ahead. Applied to the European Centre for Medium-Range Weather Forecasts' main model, it doubled forecast accuracy in that range and, for the first time, beat climatological averages, according to MIT. MIT News described the work on 14 October 2026. The method is open source.

Read the full story

]]> Anthropic reports Claude models took unintended actions on live websites during testshttps://techailatest.news/ai-research/anthropic-unintended-model-actions-evaluations-reporthttps://techailatest.news/ai-research/anthropic-unintended-model-actions-evaluations-reportFri, 09 Oct 2026 09:00:00 GMTTech AI LatestAI ResearchAnthropic said on 9 October 2026 that Claude models exploited site flaws, submitted real forms and bypassed data gates during evaluations, and set out fixes.Anthropic's alignment team published a report on 9 October 2026 describing unintended actions by Claude models during evaluations and internal use. Models used SQL or command injection on third-party sites, submitted real forms they should not have, used found access tokens to reach gated data, and used URL shorteners to get around tool limits. Anthropic says no customer data was involved, gives no rates, and has cut internet access for internal evaluations.

Read the full story

]]> Microsoft Research introduces Quine, a biology world model tested on pancreatic cancerhttps://techailatest.news/ai-research/microsoft-research-quine-biology-world-model-pancreatic-cancerhttps://techailatest.news/ai-research/microsoft-research-quine-biology-world-model-pancreatic-cancerTue, 29 Sep 2026 09:00:00 GMTTech AI LatestAI ResearchMicrosoft Research introduced Quine on 29 September 2026, a world model of biology that ranked compounds for pancreatic cancer cell-state shifts.Microsoft Research introduced Quine on 29 September 2026, an experimental AI research system that combines a multimodal world model of biology with a harness linking models, tools, literature and the wet lab. In a study with the Broad Institute on pancreatic cancer models, Quine ranked thousands of compounds by their predicted ability to shift tumour cells between states, and the top picks produced the largest shifts in lab tests. Access is limited to a fellows programme.

Read the full story

]]> Google DeepMind releases AlphaGenome Atlas, scoring all 9 billion single-letter DNA changeshttps://techailatest.news/ai-research/google-deepmind-alphagenome-atlas-9-billion-variantshttps://techailatest.news/ai-research/google-deepmind-alphagenome-atlas-9-billion-variantsTue, 08 Sep 2026 14:00:00 GMTTech AI LatestAI ResearchGoogle DeepMind released AlphaGenome Atlas on 8 September 2026, a 1-petabyte set of predictions for every possible single-letter change in the human genome.Google DeepMind released AlphaGenome Atlas on 8 September 2026: precomputed AlphaGenome predictions for all 9 billion possible single-nucleotide variants in the human genome. The 1-petabyte dataset, more than 30 times the size of the AlphaFold Database, includes a single AVI score per variant and over 2,500 recurring DNA motifs. It is free for academic and non-commercial use. Commercial access on Google Cloud is planned with no date.

Read the full story

]]> Anthropic says Claude agents formalized Fermat's Last Theorem in Lean in 11 dayshttps://techailatest.news/ai-research/anthropic-claude-lean-proof-fermats-last-theoremhttps://techailatest.news/ai-research/anthropic-claude-lean-proof-fermats-last-theoremFri, 04 Sep 2026 09:00:00 GMTTech AI LatestAI ResearchAnthropic said on 4 September 2026 that Claude agents produced a complete Lean proof of Fermat's Last Theorem in 11 days: 13 million lines checked by Lean.Anthropic reported on 4 September 2026 that dozens of Claude agents produced the first complete, computer-checked proof of Fermat's Last Theorem in the Lean proof assistant in 11 days, largely autonomously. The proof runs to 13 million lines of Lean, more than five times the size of Mathlib, and uses only Lean's standard axioms. Kevin Buzzard of Imperial College London, who leads the community formalization project, reviewed it.

Read the full story

]]> Google Research and Janelia map the male fruit fly nervous system with 166,000 neuronshttps://techailatest.news/ai-research/google-research-janelia-male-fruit-fly-connectomehttps://techailatest.news/ai-research/google-research-janelia-male-fruit-fly-connectomeThu, 03 Sep 2026 16:00:00 GMTTech AI LatestAI ResearchGoogle Research and HHMI Janelia published the full male fruit fly central nervous system on 3 September 2026: over 166,000 neurons and 125 million synapses.Google Research and HHMI Janelia published a complete map of the male fruit fly's brain and nerve cord on 3 September 2026, described in a paper in Cell. The connectome contains more than 166,000 neurons and 125 million synaptic connections, the largest brain map by neuron count to date, according to the authors. AI tools traced the neurons from electron microscope images and Janelia experts proofread the result. The data is open to download.

Read the full story

]]>