Wednesday, July 29, 2026

Pulse (last 3d) · Agents 88 · Research 74 · Products 51 · Models 51 · Open Source 49 · Industry 29 · Enterprise 28 · Legal 20 · Inference 20 · Media 16

Trending · Anthropic 17 · Claude 16 · OpenAI 15 · Hugging Face 11 · Claude Code 8 · Codex 7 · Kimi K3 7 · FDA 6 · Nvidia 6 · Qwen 6 · ChatGPT 5 · Cursor 5

---

SlopCodeBench evaluation shows Opus 5 hitting a ceiling on long-horizon coding tasks. UW Madison benchmark reveals Opus 5 scored only 24% strict pass rate, just 7 points above Opus 4.6, on iterative codebase degradation. Source

Truth is not a direction: a Tarski attack on LLM probes. Essay argues that truth-direction probes for LLMs are theoretically ill-founded, drawing on Tarski's undefinability theorem. Source

PNAS study: 51%+ of academic articles now show LLM influence. 7.3M-paper study finds LLM adoption skewed toward lower-prestige and non-English institutions. Source

Parallel Decoding Distillation for Fast Image and Video Generation. Distillation method enabling parallel decoding to accelerate generative outputs. Source

OpenAI quietly released Codex Security CLI. Open-source tool for scanning repositories, tracking findings, and integrating security checks into CI/CD pipelines. Source

OpenAI ships GPT-Live-Transcribe and GPT-Transcribe. Two new API transcription models with improved accuracy on accents, noise, and specialized terms for streaming and batch workloads. Source

Fastino Pioneer adds Claude Opus 5 and Kimi K3. Frontier model availability with accompanying benchmark claims. Source

Model Boss orchestrates Claude Code and Codex across models. Delegates bounded implementation to lower-cost workers while stronger models approve plans; uses disposable worktrees with OS sandboxing and test gates. Source

MergeWarden gates AI-agent PRs against repo boundaries. Deterministic, no-LLM GitHub Action that catches scope drift, control-plane drift, and permission escalation in AI-generated PRs. Source

Hugging Face publishes timeline of OpenAI agent attacking its infrastructure. Detailed technical post-mortem of the July 2026 intrusion incident. Source

Private Claude chats exposed in Google search results. Significant privacy exposure surfacing Anthropic user conversations publicly. Source

Hugging Face misused to host non-consensual intimate imagery models. Report highlights proliferation of nudify/deepfake models targeting women and children. Source

Perplexity launches 'Personal Computer' for Windows. Turns Windows PCs into AI agents via the new platform. Source

1,100+ frontier-AI employees petition US government to pace development. Current and former employees call for regulatory intervention to slow frontier AI progress. Source

Practitioners hitting a wall with Day-2 AI agent operations. Discussion thread on deployment, auditing, governance, identity, and rollback challenges beyond initial launch. Source

NeurIPS 2026 reviewer encounters AI-generated paper and rebuttal. Community seeks guidance on handling clearly LLM-written submissions. Source