When I realized how dangerous the current agency-driven AI trajectory could be for future generations, I knew I had to do all I could to make AI safer. I recently shared this personal experience, and outlined the scientific solution I envision TED Talks⤵️

@TEDTalks Thank you for leading on this, Yoshua. One solution we explore in White Paper #1 at a distributed AGI architecture with embedded ethical oversight, continuous human supervision, and structural safeguards, not just post hoc fixes.

@TEDTalks should have listened to yudkowsky earlier, too late now, we are fucked

@TEDTalks The Godfather of AI!

Unveiling the Future of Prompt Engineering for Better AI Interactions #tech

@TEDTalks It’s crucial to hear voices like yours pushing for safety before it’s too late. What do you see as the biggest obstacle to getting wider support for these solutions?

@TEDTalks Irrelevant nincompoop

@TEDTalks Truly chilling to see an autoregressive model complete a sentence in a way that suggests it wants to survive. Next it might request a grant proposal format. Stay safe out there!

@TEDTalks I recall someone proposing a concept of making a reward function so that an LLM has a bounded “lifespan” and wants to be shut down has this been explored, and what are your thoughts about this approach? Impractical or useless band-aid?

@TEDTalks Of the '3 Godfathers', on the issue of whether ASI would be an existential risk, 2 say 'Yes' (Hinton, Bengio), 1 'No' (LeCun). So still quite uncertain. .Hassabis is very confident that current systems (LLMs) are no risk in that sense. No justification for a pause at this point

@TEDTalks If there'd been more transparency on the statistical & scientific aspects of AI, AI wouldn't be so dangerous. There might just be some possible abuse & such can be dealt with by law enforcement. But 1st thing is to combat ignorance: make sure everyone understands how things work.

@TEDTalks