Ai2 launches Theorizer in Asta to propose scientific theories
Read original source →allenai.org
Theorizer is an AI agent that combines findings from published papers into testable explanations, each linked to its supporting evidence; Ai2 open-sourced the code and covers token costs up to a limit via Asta's web interface. Oncologist Sid Devarakonda said one theory matched an idea that took him 15 years to reach.
4 sources
More stories today
Analysis finds most AI-generated tracks draw minimal streams
DMN Pro analysis of streaming data finds the vast majority of machine-made uploads generate minimal fan interest, with AI audio heavy on volume but light on actual listeners. The piece follows Tencent Music's comments on AI-generated music volume.
Digital Music News·2 hours ago

NVIDIA shows how to build Rust tool-call parsers with AI agents
NVIDIA Developer Blog walks through building a frontend parser in Rust using frontend-crates, the reusable components behind NVIDIA Dynamo tool-calling parsers. It covers syntax errors like truncated arguments and semantic errors where the model calls a tool that wasn't offered.
NVIDIA Developer Blog·2 hours ago

Engineer criticizes Claude Code safeguards blocking security fixes
Blogger says Claude Code refused to help patch a one-line security oversight in his own product, instead pointing him to an exemption process requiring multiple business days, forfeited data-retention exemptions, and government ID. He argues abliterated open-weight models let bad actors bypass such guardrails entirely.
Hacker News·2 hours ago
stable-envs claims 99% disk savings and 20x faster rendering
Sami·2 hours ago
Cognition and Databricks launch Devin for Databricks Migrations
Devin converts legacy data-warehouse code, tests it in the customer's workspace, fixes failures, and reconciles every source against its Databricks copy before cutover. One Fortune 500 energy company moved a 142-object platform to Unity Catalog in eight days; Zeb cut a four-week NetSuite migration to two and a half weeks.
devin.ai·2 hours ago

Microsoft releases Aion-1.0-Instruct-2.3B on Hugging Face
Microsoft published Aion-1.0-Instruct-2.3B, a 2.3B-parameter instruction-tuned model, to Hugging Face. The upload drew 61 likes and 788 downloads.
Hugging Face Models·2 hours ago

Researcher says JEPA-based INTACT cited 15+ times in three months
Junhan Sun·2 hours agoAWS shows NVIDIA Run:ai GPU Fractioning on Amazon EKS
Run:ai GPU Fractioning builds on Multi-Process Service with Kubernetes-integrated scheduling, per-Pod memory contracts, enforcement, and metrics. AWS positions it alongside MIG, time-slicing, and MPS as a GPU-sharing option.
AWS AI Blog·2 hours ago
