AI that matters, from the architect's desk. Curated and engineered by Saaket Varma, PhD — no hype, just signal.
HOW TO READ THIS Top to bottom: Lightricks ships LTX-2, studio video moves from closed vaults to open, the model syncs video with sound, and the result is studio-grade output anyone can access.
Lightricks open-sourced LTX-2, a generative model that produces synchronized audio and video, and shipped the inference code plus a LoRA trainer alongside it. That combination — open weights you can run on a single GPU, plus a trainer to fine-tune on your own footage — drops studio-grade generative video out of closed platforms like Sora and Veo and into anyone's hands. The strategic point isn't the demo reels; it's that the capability is now forkable, auditable, and yours to specialize. Clone the repo, run it on your own GPU, and train a LoRA on your own footage before this layer gets gatekept again.
HOW TO READ THIS Read top to bottom: the laureate leaves DeepMind, arrives at Anthropic, and joins its research core.
When the Nobel-winning AlphaFold scientist walks from DeepMind to a rival — and he isn't the only one — that's elite talent voting on where the frontier momentum actually is.
HOW TO READ THIS Read downward from four music datasets through artist search to track membership, following [The Atlantic’s investigation](https://www.theatlantic.com/technology/2026/06/ai-music-generators-suno-google-udio/687485/).
A public, searchable look at exactly whose work fed the models — concrete ammunition for the copyright fights that will shape how the next generation gets built.
HOW TO READ THIS Read top to bottom: agents that forget get a Cognee memory layer, which lets them recall facts across sessions, earning 18.5K GitHub stars.
An open-source, self-hosted memory layer that gives agents persistent recall across sessions — the missing piece between a chatbot and an assistant that carries context day to day.
Stanford's LLM research agent turns a topic into a full-length, citation-backed report — knowledge curation, not just generation.
An agent development environment for running a fleet of parallel coding agents under your own subscription.
754 structured cybersecurity skills for AI agents, mapped to MITRE ATT&CK, NIST CSF 2.0, and MITRE ATLAS.
Open-source voice studio — clone, dictate, and generate speech without committing to a closed API.
Open-source AI + no-code platform for standing up internal business systems fast.
Interconnects Nathan Lambert's case against banning open-source AI — timely as Washington pressure on labs mounts.
Latent Space GLM-5.2 is passing vibe checks against GPT, and Z.ai is forecasting an open Fable-class model by December.
The Sequence Last week in AI: a $60B Cursor deal, Google's brain drain, and Midjourney's body scanner.