Sunday, July 19, 2026

Pulse (last 3d) · Agents 66 · Research 58 · Models 36 · Products 31 · Industry 27 · Open Source 23 · Enterprise 18 · Releases 17 · Legal 15 · Infra 9

Trending · OpenAI 12 · Kimi K3 10 · Anthropic 8 · Claude 8 · DeepSeek 7 · Apple 4 · Artificial Analysis 4 · ChatGPT 4 · Fable 5 4 · Google 4 · Inkling 4 · Moonshot AI 4

---

Biohub ESMC/ESMFold2: open protein binder design outperforming AlphaFold 3. Biohub release of an AI drug discovery platform for protein binder design with wet-lab validation, beating AlphaFold 3 on antibody-antigen tasks. Source

GPT-5.6 closed a 30-year gap in convex optimization. Discussion of GPT-5.6 producing a proof that closes a ~30-year complexity gap in convex optimization bounds. Source

General-purpose LLMs outperform FDA-cleared clinical AI. Nature Medicine study shows general LLMs beat FDA-cleared clinical AI on physician queries, exposing a regulatory gap. Source

AI-Designed Synthetic CRISPR Nucleases (SynTnpB) Exceed Wild-Type. Science paper reports AI-designed CRISPR nucleases achieve editing efficiency exceeding wild-type variants. Source

Claude Code now runs on Bun's Rust port. Binary inspection confirms Claude Code uses Bun v1.4.0 (Rust), yielding ~10% faster Linux startup. Source

Sandbox SDK. Open-source SDK for spinning up sandboxes across Vercel, E2B, Daytona, and Upstash with one unified syntax. Source

SolidState HQ. Open-source runtime separating probabilistic reasoning from deterministic execution for safe, controllable AI agents. Source

Harmonist. Portable AI agent orchestration framework with mechanical protocol enforcement and ~186 bundled agents. Source

Kimi: Threat or menace?. TechCrunch frames Moonshot's Kimi as a competitive threat to closed-source frontier labs. Source

Have we reached AI's 'good enough' era?. Discussion argues open-weight models like Kimi K3 and Fable signal diminishing returns for frontier labs. Source

Did blatant AI slop win a $25K DeepMind/Kaggle prize?. Critique alleges the 'Cognitive Abilities' grand prize went to a low-quality, AI-slop-laden submission. Source

Dave Eggers tells OpenAI staff ChatGPT is 'silencing a generation'. Author Dave Eggers confronted OpenAI staff over ChatGPT's cultural impact. Source

LLM debate induces 'persuasive hallucination'. Author finds LLM-vs-LLM debate setups produce fabricated citations used as rhetorical weapons, beyond simple sycophancy. Source

Cut RAG latency from 90s to 4s without touching the model. Engineering write-up shows a 95% cost and latency reduction from fixing retrieval, not the LLM. Source

Deep learning for scRNA-seq: survey of 25 methods. Walkthrough and summary tables covering 25 deep-learning methods for single-cell RNA-seq analysis. Source