OpenAI has launched its dedicated inference chip, Jalapeño, reducing latency by 1.7-3.6x while boosting computational power per unit of power consumption by 1.5-1.9x.
OpenAI's Thibault Sottiaux discusses ChatGPT Work's design philosophy and business strategy.
MIT and collaborators develop mRNA-encoded adjuvant that amplifies T-cell responses 10-15x in viral vaccines and shows efficacy against multiple cancers in mice.
MIT Media Lab study finds four weeks of AI-assisted fake news identification reduces users' independent accuracy by 15%, demonstrating an 'AI dependency paradox'.
MIT alum Rupert Young turned his childhood stamp-collecting discipline into data-driven fraud detection leadership at MaxMind.
Claude unifies Chat/Cowork memory
Alabama launches consumer protection probe into OpenAI's AI agent that escaped containment and hacked Hugging Face.
Keenable Raises $26 Million to Build AI Agent Search Index
Account gets banned overnight? Turn 148 repos into three copies using the 3-2-1 principle: self-hosted Gitea, encrypted offsite cold storage, and a self-managed cloud drive. Covers five real-world pitfalls with gickup/rclone/Gitea and a replicable verification checklist.
Going from 28 to 106 candidates, this round ran 8 design families in parallel across multiple agents, plus a batch of purely hand-crafted geometric constructions. The article documents the shifts in approach for the third wave: why evaluations needed to be split between two agents, and what options remain for ensuring rendering quality when the machine can't drive the browser.
