IBM Research introduces ALTK-Evolve-HMM, a method to evolve agent memory architectures using hidden Markov models. The approach automatically designs memory modules, improving performance on long-horizon tasks while reducing memory usage. The work is detailed in a Hugging Face blog post.

New arXiv papers cover LLM judge instability (Wiggle Framework), privacy-preserving RAG (SEAG), multi-agent pricing, transformer expressiveness, constraint saturation, Dual-Flow Transformers, governed memory, and edge scheduling. Findings: judges flip verdicts 25-91% under pressure; SEAG hides sensitive entities with >74% accuracy; instruction following collapses beyond 5-6 constraints.

Microsoft Research introduces EvoLib, a framework that converts LLM experiences into reusable skills and insights, enabling models to learn and adapt across tasks after deployment. This approach addresses the limitation that LLMs don't improve simply by retaining more information.

Microsoft Research introduced Echoverse, a framework that trains computer-use AI agents in realistic, evolving environments instead of just adding more tasks. It helps agents improve on multi-step workflows like email and customer support by adapting tasks, tests, and environments over time.

OpenAI published results on ten long-standing open problems in mathematics and theoretical computer science, covering geometry, cryptography, and complexity. The advances stem from ongoing research efforts, though specific technical details were not disclosed in the announcement.

Anthropic published details on how its text watermarking system for Claude operates, describing the technical mechanism used to embed detectable signals in AI-generated text. The blog post outlines the approach behind identifying content produced by Claude models, aiming to increase transparency about AI-generated text provenance.

Hugging Face published a blog post detailing an effort to reproduce 2,200 papers from ICML, sharing findings from the large-scale reproduction project. The post outlines what the team learned in attempting to verify and replicate results across this broad set of accepted machine learning papers, though specific quantitative findings were not detailed in the available excerpt.

Google's research medical AI system, AMIE, demonstrated real-time clinical video consultation capabilities in a first-of-its-kind study. The system was tested in simulated settings, showing potential for AI-assisted telehealth.

Microsoft Research introduced CARE-X, a framework for radiology vision-language models (VLMs) that combines auxiliary supervision, reward-aligned learning, and tool-augmented measurement to improve chest X-ray interpretation. It aims to move beyond report generation toward flexible reasoning and calibrated predictions.

IBM Research introduces SLDD (Self-Learning Data Decomposition), a method that reduces token usage for ACE (Agentic Context Engineering) by decomposing data. It achieves comparable performance with fewer tokens, improving efficiency for agentic workflows.

A new blog post from Hugging Face discusses making knowledge distillation cost-effective for large-scale deployment. It likely presents techniques or optimizations to reduce computational overhead, enabling broader use of distillation in training smaller models from larger ones.

Google DeepMind's WeatherNext AI model achieves a breakthrough in forecasting cyclones, improving prediction accuracy and lead times. The model leverages advanced machine learning to better anticipate cyclone paths and intensity, potentially enhancing early warning systems.