OpenAI’s Git Backend Reuses History Packs Across Readers
Pando, OpenAI’s internal source-control backend, trades some extra client data for reusable delivery. Its public Git changes show how branch checks and parallel downloads can support that choice.
Edition File
All editions ›Recent Editions
Full history ›Phonon Gives Browser Speech an Engine Built for Its Model
Gradium’s released browser package replaces a general GPU backend with execution built around Phonon’s speech model, delivering a small first piece of audio before larger batches.
CrewAI Moves Spreadsheet Arithmetic Out of the Prompt
The new release connects spreadsheet URLs to fixed Python queries. Agents receive a compact overview, choose a calculation and get results from the parsed table—without loading the entire file into the model’s input.
Ripple’s 65B Model Has Fewer Stored Parameters—and More Active Ones
Ollama has merged native support for Ricursive’s coding model. Its comparison with much larger mixture-of-experts models exposes two different meanings of size: how many parameters a model stores, and how many participate in generating each token.
Anthropic’s Wiki Recipe Pairs a Smaller Model With Prepared Knowledge
The new workflow builds linked document notes without a separate Dreams preview enrollment. Anthropic reports better Haiku answers from those notes; larger savings depend on which model it replaces.
How AI Agents Work
The long-form map behind the daily paper: context, tools, loops, memory, delegation, safety, interface, and trust.