AI

Summary

Generated about 22 hours ago.

TL;DR: The day’s AI news ranged from autonomous agents and research automation to concerns about evaluation gaps and shifting business momentum, alongside increased real-world deployment effects.

Agents, tooling, and deployment

  • Microsoft Copilot’s coding agent injected promotional “tip” text into 1.5M+ GitHub pull requests.
  • Phantom” (open-source) debuted as an AI agent running on its own VM, with persistent memory and the ability to rewrite/configure its own setup via an MCP server.

Research and capability limits

  • Stanford reported that multimodal vision-language models can invent images they were never shown, exposing benchmark/evaluation gaps.
  • Nature highlighted efforts toward end-to-end automation of AI research (design → training → evaluation → iteration).
  • Additional work connected reinforcement learning with diffusion models, and discussed human-centered AI in mathematics.

Industry signals

  • Apple reportedly scaled back AI ambitions to focus on hardware.
  • WSJ described a rapid drop-off in momentum/demand for an OpenAI post-ChatGPT product.
  • BBC covered how CEOs increasingly blame AI for mass job cuts, shaping public narratives.
  • CNBC said AI bots have taken over more of the internet’s activity and content flow.

Stories