Pulse (last 3d) · Agents 64 · Products 63 · Research 59 · Models 48 · Enterprise 31 · Industry 26 · Legal 21 · Media 19 · Inference 18 · Open Source 12

Trending · OpenAI 18 · Claude 15 · ChatGPT 11 · Gemini 7 · Claude Code 6 · Codex 6 · Tempus 6 · llama.cpp 6 · Slack 5 · Astra 4 · Perplexity 4 · AMD 3


DeepMind's hurricane breakthrough has surprised weather scientists. DeepMind's AI-based hurricane forecasting model has produced predictions that surprised meteorologists with their accuracy, reportedly buying forecasters an extra day of lead time. Source

Oncoformer: Multimodal Clinical Transformer on 3.67M Patients. Liu et al. publish a clinical-genomic foundation model trained on 3.67M patients in Cell (July 2026), targeting oncological prediction. Source

Deterministic screen-activity compilation for agent memory. A zero-model pipeline compiles passively captured screen activity into typed "activity frames" that agents read as memory, achieving 86× compression and 98.4% QA accuracy vs. 66–80% for LLM summaries. Source

DASH: Temporal Coefficient Allocation for On-Policy Self-Distillation. DASH recovers order-dependent trajectory coefficients dropped by vanilla OPSD via adaptive propagation gates, with zero additional cost. Source

AI in Drug Discovery: A Decade of Limited Clinical Impact. Bender et al. (Nat Rev Drug Discov, Aug 2026) review AI's clinical impact on drug discovery and propose a path forward. Source

Auto mode is now the default in Claude Code for Pro, Max, and Team plans. Anthropic sets Auto mode as the Claude Code default starting August 14, 2026, signaling confidence in its safety guardrails. Source

ToolAhead. A local-only, MIT-licensed tool that pre-runs agents' recurring tool calls and serves results only if the exact call and workspace still match; works with Codex CLI, Claude Code, and MCP. Source

Ordewell. A free, Apache-2.0 planner that turns one goal into an ordered plan of coding-agent tasks with per-task runner/model/effort; you edit the plan before any tokens are spent. Source

Ferrox: Pure-Rust Inference Engine Reaches llama.cpp Parity on Apple Metal. A from-scratch Rust GGUF engine (no llama.cpp/ggml bindings) matches llama.cpp within 5% on Llama-3.1-8B Metal and beats it 20–48% on sub-1B models. Source

Meta debuts first AI coding agent to take on Anthropic and OpenAI. Meta enters the AI coding-agent space with "Muse Code," its first product aimed at Anthropic and OpenAI. Source

ByteDance trains massive AI model in bid to rival Anthropic. ByteDance is training a frontier-scale model intended to compete with Anthropic's leading systems. Source

OpenAI acquires presentation startup NextSlide. OpenAI acquires NextSlide, a startup building AI-powered presentation tools. Source

Now we have a timeline of the OpenAI accidental attack against Hugging Face. Simon Willison assembles a timeline of the recent OpenAI/Hugging Face incident, with Zvi Mowshowitz's explainer providing context. Source

So AI has now designed actual viruses that work. Reported research indicates AI tools have designed viruses that function in the lab, raising biosecurity concerns. Source

Real-Time Conversational Agents (RTCA) Workshop @ NeurIPS 2026. Submissions open for the RTCA workshop at NeurIPS 2026 (full-duplex voice and streaming dialogue systems); deadline Aug 29 AoE. Source

Building a zero-dependency C inference engine for BitNet. Author details a zero-dependency C inference engine for BitNet (1.58-bit) LLMs, reporting 36 tok/s on a Xeon CPU. Source

NeurIPS AI Assisted Review authors/reviewers?. NeurIPS reviewers report mixed experience with AI-assisted reviewing, including double-blind violations and superficial reviews. Source