tszzl: AI safety is solvable, maybe soon, because models are fully legible programs
tszzl · x · 2026-10-13
In an exchange with Rohit Krishnan, investor tszzl argues that unlike companies or people, AI is both more dangerous and more legible — every piece of its internals can be read since it's all computer programs. He believes the safety problem can be solved, possibly soon.
More from AGI Musings
- OpenAI researcher: progress will soon be data-bottlenecked, not intelligence-bottlenecked — giffmana · 2026-10-13
- Researcher: Artificial Life will be a key research direction as agentic AI gains self-replication and open-endedness — burny_tech · 2026-10-13
- AI-generated ads quietly take over: half of served IG and YT ads now look AI-made — eptwts · 2026-10-13
- Daniel Jeffries: only ~2% can actually use AI agents, mass adoption is 10-20 years away — Dan_Jeffries1 · 2026-10-13
- From chatbots to agents to personal AI assistants: the endgame is here — koltregaskes · 2026-10-13
- AI content can be useful and still make the internet worse: no POV, no skin in the game — Certain-Walrus5060 · 2026-10-13