AI

Daniel Kokotajlo and the AI Futures Project released an ambitious 15-year roadmap for managing advanced AI development through international cooperation. Here's what HN thinks about it.

A new experimental technology encodes messages in video using motion-based steganography, exploiting how AI models process video as individual frames rather than continuous motion.

pgvector, Pinecone, Qdrant, Weaviate, Chroma, Milvus, and Turbopuffer compared on hosting model, filtering, scale, and cost for RAG.

An 82M parameter text-to-speech model that runs on CPU and produces high-quality speech across multiple languages - no cloud APIs or GPU required.

Mistral's new 8B parameter model enables robots to navigate complex environments using only a camera and natural language commands. Here's what it does, how it works, and what the benchmarks actually mean.

A CS student built 30papers.com to make Ilya's legendary ML reading list more accessible. HN has thoughts on the source, the format, and why compression equals intelligence.

IEEE Spectrum reports on pharmaceutical AI running on handheld devices. HN debates emergency kits, domain-specific models, and whether AGI will emerge from scaling or specialization.

A new study from Dartmouth measures the impact of an AI tutoring platform on introductory statistics performance. Full engagement with the system correlated with significant exam score improvements, though selection bias remains a key limitation.

Comparing LLMs by token pricing alone can lead you to choose worse, more expensive models. Cost per task tells the real story.

A new paper proposes inverting traditional agent architecture - making the append-only event log the source of truth, not an afterthought. HN debates whether this is novel or just CQRS with extra steps.

A companion guide to the Agents 101 video: a behind-the-scenes walkthrough of building and deploying AI agents fast on Vercel, the agentic infrastructure stack. Here is the map of what to learn and where to go next.

DeepReinforce AI released Ornith-1.0, a family of open-source coding models claiming self-improvement. The HN thread reveals a mix of skepticism and genuine interest - here is what the model actually does and whether the hype holds up.

A developer fed 266MB of DICOM MRI data to Claude Code Opus for a second opinion on a shoulder diagnosis. The AI disagreed with the doctor. HN radiologists weighed in.

Semgrep's security research team benchmarked LLMs on IDOR vulnerability detection. The open-weight GLM 5.2 beat Claude Code by 7 points at roughly one-sixth the cost.

Filippo Valsorda argues that LLMs have ended the era of treating security researchers with kid gloves. When anyone can discover vulnerabilities with an AI, the old coordinated disclosure model breaks down.

Baidu releases Unlimited OCR, an open-source vision-language model that parses 100+ page documents in a single pass without memory blowup. Here's what developers need to know.

The new wrangler deploy --temporary flag creates ephemeral Cloudflare accounts for AI agents. 60-minute deployments, no OAuth, no browser - just deploy and claim later.

Modern LLMs now use MoE routing, mixed attention variants, and fused vision encoders. The simple transformer stack is gone - here's what replaced it and why it matters for developers.

The Transformer co-creator leaves Google DeepMind for OpenAI just two years after Google paid $2.7 billion to bring him back from Character.AI.

Anthropic's first generally available Mythos-class model, released June 9, 2026. 1M context, 128K max output, $10/$50 per million tokens. Built for long-horizon agentic work.