Ilya Sutskever’s guess for AI: an internal value function that enables lifelong learning
FinanceYF5 · x · 2026-07-29
He sketches a loop of experience → internal judgment → self-correction → lifelong learning → transfer → repeat.
The underlying guess is that humans don’t wait until a 10,000-step task ends to know they failed; emotions and intuition provide ongoing feedback. He speculates that the corresponding AI ingredient may be some kind of internal value function.
More from AGI Musings
- AI doom probability is probably closer to 70%, says the author — AaronBergman18 · 2026-07-29
- Researcher Warns of Emerging 'AI Doomer Religion' Treating CEOs as Saviors — davidpattersonx · 2026-07-29
- AI is turning copyright into something that belongs to no one — LiteBlackEagle · 2026-07-29
- Geoffrey Irving says AI coordination may matter as much as acceleration — geoffreyirving · 2026-07-29
- Christoph Szegedy warns bad actors could gain a decisive advantage from faster AI — ChrSzegedy · 2026-07-29
- Open-source AGI may leave frontier labs more exposed at the API layer — pzakin · 2026-07-29