Science

AI research and science – new models, studies and breakthroughs from labs and universities, clearly put in context.

Research
GPT-5.6 Disproves Twenty-Year-Old Statistical Assumption in 90 Minutes

Statistician Edgar Dobriban of the University of Pennsylvania has used the language model GPT-5.6 Sol Pro to disprove an assumption about the Benjamini-Hochberg procedure that had been considered…
Research
Anthropic Study: Claude Responds Warmer in Hindi Than in Russian

Anthropic has published a study on the values of its AI model Claude, which finds significant differences depending on language and model version. In Hindi and Arabic, the model responds warmer and…
Research
Soofi S: German Consortium Releases Open AI Model

A German research consortium has released Soofi S, an open language model trained specifically for German and English. According to its developers, the 31.6-billion-parameter model surpasses the open…
Research
Google Trains SensorFM on 1 Trillion Minutes of Wearable Data

Google Research has introduced SensorFM, an AI model that learns general patterns of human health from wearable data. It is built on more than a trillion minutes of sensor readings from five million…
Research
Oak Lab: Sutton Builds AI Agents That Learn on the Job

Reinforcement learning pioneer Richard Sutton has founded the startup Oak Lab together with his former doctoral student Khurram Javed. The Canada-based company wants to build AI agents that keep…
Research
AgenticSTS: New AI Memory System Doubles Win Rate in Card Game

An international research team has introduced AgenticSTS, a new memory system for AI agents, tested in the card game Slay the Spire 2. Five separate memory layers replace the previously common…
Research
Orca: BAAI World Model Matches Robotics Without Action Labels

The Beijing research institute Beijing Academy of Artificial Intelligence (BAAI) has introduced a novel world model called Orca, which derives text, images, and robot commands from a single internal…
Research
GPT-5.6 Sol Ultra proves math conjecture – experts still examining

OpenAI published a proof of the Cycle Double Cover Conjecture on Friday, a graph theory problem that has remained unsolved for fifty years. The language model GPT-5.6 Sol Ultra generated the…
Research
arXiv:2607.08573 – SHAP-weighted Fusion for Emotion Recognition

Note: This paper describes a current, yet to be peer-reviewed arXiv preprint (as of July 9, 2026). The reported results come from the authors' own publication and have not yet been externally…
Research
arXiv:2607.08625 – How communication style influences AI triage

Note: This contribution summarizes a current, yet-to-be-peer-reviewed arXiv preprint. The results are preliminary.
Research
arXiv:2607.08652 – Formal Mechanisms for Stable AI Agent Markets

Note: This article summarizes a current, not-yet-peer-reviewed arXiv preprint. The findings are preliminary and come from a single simulation study.
Research
arXiv:2607.08716 – Proactive Memory Agent for Long-Horizon Agents

Note: This is the summary of a current, (not yet) peer-reviewed arXiv preprint. The results described come from a single study and have not yet been independently verified.
Research
arXiv:2607.08734 – The Illusion of Equivalency in Quantization

Note: The work discussed here is an arXiv preprint. It has not yet undergone a regular peer-review process; the results are not yet independently confirmed.
Research
arXiv:2607.08740 – Semantic Memory for LLM Workflows

Note: The work discussed here is an arXiv preprint. It has not yet undergone a regular peer review process; the concepts presented have not been independently verified.
Research
arXiv:2607.08745 – VQA-Benchmark for Accident Scenes via Dashcam

A research team led by Siddharth Damodharan, Radhika Gupta, Ali Alshami, Ryan Rabinowitz, and Jugal Kalita has introduced a new benchmark called AUTOPILOT VQA, which tests how well multimodal AI…
Research
arXiv:2607.08758 – Ideas Have Genomes: Idea Trees for AI

Note: The work discussed here is an arXiv preprint. It has not yet undergone a regular peer review process; the reported results are independently unverified.
Research
arXiv:2607.08748 – AI Learning Assistants in Higher Education

Note: This paper describes a (not yet) peer-reviewed arXiv preprint. The reported results come from the authors themselves and are currently unverified.
Research
arXiv:2607.07321 – EvoSOP: SOPs for Self-Learning LLM Agents

Note: This post describes a (not yet) peer-reviewed arXiv preprint. The reported results come from the authors themselves and are currently unverified.
Research
arXiv:2607.08602 – Clinical AI Model for Liver Cancer Therapy

Note: This contribution summarizes a current, yet-to-be-peer-reviewed arXiv preprint. The results are preliminary and stem from a single study conducted by the authors.
Research
arXiv:2607.08681 – SolarChain-Eval: AI Agents in the Energy Market

This is the summary of a current arXiv preprint (as of July 9, 2026). The paper has not yet been peer-reviewed; the results are preliminary.
Research
OpenAI: SWE-Bench Pro is about 30 percent faulty

Two days after the broad rollout of GPT-5.6, OpenAI has retracted its own recommendation: The coding benchmark SWE-Bench Pro, which the company had only proposed in February 2026 as a yardstick for…
Research
arXiv:2606.15943 – Graphical Models for Generative AI Software Systems

Note: This paper describes a (yet) non-peer-reviewed preprint on arXiv. The mentioned contents have not been independently verified by the scientific community so far.
Research
arXiv:2606.15954 – Green SARC: Budget Governance for AI Agents

Note: This post describes a (still) non-peer-reviewed preprint on arXiv. The results mentioned have not yet been independently verified by the scientific community.
Research
arXiv:2606.15956 – TDV: Self-Supervised Vision Without Assumptions

Note: This post describes a (not yet) peer-reviewed preprint on arXiv. The results mentioned have not yet been independently verified by the scientific community.
Research
arXiv:2606.15963 – PreLort: LoRA for Federated Fine-Tuning

Note: This paper describes a (not yet) peer-reviewed preprint on arXiv. The results mentioned have not been independently verified by the scientific community so far.
Research
arXiv:2606.15959 – Lossy Compression for AI Surrogates

Note: This post describes a current arXiv preprint (arXiv:2606.15959, submitted on June 14, 2026). A preprint has not yet been peer-reviewed by independent experts; the results presented are solely…