OpenAI published the first measured results for Jalapeño, its custom inference chip; Cognition wrote up the chat-renderer rebuild that makes hundred-thousand-message Devin sessions usable; GitHub shared hard-won lessons from evaluating an LLM system for secret scanning; and Simon Willison flagged EVE Online’s long-overdue move off Stackless Python 2.
From the labs
- Jalapeño’s first results show industry-leading speed and efficiency in AI inference — OpenAI
- How Devin rebuilt its renderer — Windsurf
- How to evaluate LLMs before production — GitHub
Worth your time
- EVE Online: The Move to Python 3 Begins! — Simon Willison