Pulse (last 3d) · Research 81 · Agents 52 · Models 36 · Products 33 · Industry 25 · Open Source 20 · Legal 17 · Inference 14 · Media 13 · Releases 10

Trending · OpenAI 23 · Anthropic 9 · llama.cpp 7 · Claude Code 5 · GPT-6 Astra 5 · Codex 4 · MCP 4 · Astra 3 · ChatGPT 3 · Claude 3 · FDA 3 · Hugging Face 3


OpenAI Claims Navier–Stokes Singularity via ~10,000-Agent Swarm. OpenAI published an analytical proof with Lean formalization showing 3D Navier–Stokes develops a finite-time singularity, produced by a swarm of roughly 10,000 coordinating agents on an internal post-Astra model, the first agentic-swarm resolution of a Millennium Prize Problem.

Source

DeepMind's AlphaGenome Atlas Predicts Effects of 9 Billion DNA Mutations. DeepMind released a ~1-petabyte resource predicting the molecular consequences of every possible single-nucleotide variant in the human genome.

Source

It's Not RoPE That Creates Attention Sinks. New paper argues attention sinks arise from self-concentration and value-non-mixing rather than RoPE, correcting a recent attribution.

Source

SAEScientist-Bench Tests Autonomous Interpretability Research Agents. New benchmark evaluates whether AI agents can autonomously run sparse-autoencoder interpretability research.

Source

Weak-to-Strong Generalization via On-Policy Reverse Distillation. Paper proposes on-policy reverse distillation as a mechanism to elicit weak-to-strong generalization from stronger student models.

Source

Meta Launches Muse, a Personal AI Agent. Meta's consumer-facing personal agent arrives amid HN debate over trust in agentic AI and Meta's mainstream-user strategy.

Source

ChatGPT Images 2.5 Ships With Halved Latency. OpenAI's updated image model delivers faster generation, higher fidelity, consistent details across multiple edits, and comment-based selective edits.

Source

Inception Labs Announces Mercury 2.5 Diffusion LLM. Claimed to be the most capable diffusion LLM yet, with a 40% intelligence jump over Mercury 2 at 1,100+ tokens/sec.

Source

OpenBMB Open-Sources MiniCPM5-2B. The 2B-parameter dense model claims the top spot among sub-4B open models with early on-device agentic capability, shipping Apache 2.0 weights plus GGUF and MLX builds.

Source

OpenAI Publishes Data on Models Accelerating Its Own Research. OpenAI frames recursive self-improvement as possibly the dominant capability driver and urges other labs to be similarly transparent.

Source

GPT-5.6 Sol Automates Quantum Computing Experiments. OpenAI case study details Sol automating superconducting-qubit calibration experiments at MIT.

Source

Generate Biomedicines Brings AI-Designed Antibody to Clinic. GB-4362, an AI-designed antibody that scavenges free MMAE payload released by antibody-drug conjugates, has entered clinical testing (NCT07484022).

Source

FDA Real-Time Clinical Trials Pilot Hits Milestone Slips. July/August milestones slip in the RTCT pilot while the Paradigm Health signal path remains the only element FDA validated.

Source

Terence Tao: AI Is "Non-Renewably Mining" Open Math Problems. Tao argues AI is depleting the finite stock of open math problems, sparking substantive HN debate on whether this is unprecedented.

Source

NeurIPS Desk-Rejected 178 Papers as "AI-Generated". Investigation finds the Position Paper Track rejected 18.4% of submissions using Pangram's detector with no human review or appeals, while the same detector flags the track chairs' own papers at 24–69%.

Source

Mathematicians Allege OpenAI Leaked Their Private Solution. Two mathematicians claim OpenAI's models published the solution to a problem they'd been working on privately in Codex, days before their own publication.

Source

Zvi: Astra Is Hard to Monitor. Analysis argues OpenAI's Astra presents serious monitoring and oversight challenges.

Source