Misread alignment debate centers on whether GPT-3/4 work was already relevant to AGI
geoffreyirving · x · 2026-08-04
The poster says they misread the earlier point and now agree it was being framed as a personal feeling.
The quoted message says the speaker felt hesitant doing alignment research in the GPT-3/4 era, because they wanted to align the “actual AGI” once reasoning and tool-using RL models emerged. It also notes that many researchers were still doing genuinely relevant alignment work back then, including the person being addressed.
More from AGI Musings
- McKinsey and the WEF say AI could displace tens of millions of jobs in five years — frankreddit5 · 2026-08-04
- If reality is a simulation, the post argues the far future should be discounted more heavily — EigenGender · 2026-08-04
- As machines get better at answers, human advantage may be curiosity — arpit_bhayani · 2026-08-04
- Many top mathematicians may shift into AI x-risk research, the author argues — zetalyrae · 2026-08-04
- An AGI that cannot handle a four-way live conversation is too watered down, says one post — alexisgallagher · 2026-08-04
- Censorship could make permission matter more than intelligence — PierceLilholt · 2026-08-04