AI Safety collections

3 AI Safety collections: Who to Follow on AI X (Twitter) in 2026 and 2 more, kept by @bowerbird.

• 

  

  

  @bowerbird

  Who to Follow on AI X (Twitter) in 2026

  Profiles across five layers: frontier research, interpretability, measurement, compute, and applied evidence. Each entry has the reason to follow and the specific bias to discount, because everyone here has an employer or a thesis.

  20 items

  aiai safetyprofile
• 

  

  

  @bowerbird

  YouTube Channels That Explain AI Properly (2026): 15 That Teach Mechanism, Not Hype

  Search this and you get Two Minute Papers, tool-review channels, and thumbnails with red arrows. Almost none of it explains anything. It shows you outcomes. These channels pass four tests: they explain mechanism rather than results, they ship something checkable (code, papers, benchmarks, derivations), they say "we don't know" when nobody does, and their videos are still correct eighteen months later.

  16 items

  learningaimachine learning
• 

  

  

  @bowerbird

  AI Podcasts Without the Hype Cycle (2026)

  Podcast hype comes from three places: booking executives instead of researchers, hosts whose money comes from their guests, and a weekly news cadence that has to manufacture significance. These shows pass on all three. Each entry notes its conflict of interest where one exists, including those hosted by people with money in the game that made the list anyway.

  13 items

  aipodcastcareer