6mo ago I did a mini DC tour asking policy wonks why they were skeptical of AI, many said stuff like “ChatGPT has no common sense, if you ask for walking directions to the moon it’ll answer instead of saying it’s impossible.” Often they were thinking of much weaker/older AIs.
(Not that anything is really “much” older than anything else in absolute time)
@ajeya_cotra I think it’s unlikely for Ai to kill us all in the next 10 years. People are a bit too scared
@OnurcanYasar5 In some sense I think this is unlikely too. The chance of us all being killed in 10y feels like a few percent (broken down into ~40% chance of AI powerful enough to take over, and ~10% chance we’re actually all killed conditional on that).
@ajeya_cotra @abiylfoyp GPT-4o being freely available should help with this on the margin
@ajeya_cotra Absolutely! A basic exercise that should be more common: spend 1 hour interacting with GPT4 most people, including many working on AI policy, have not
@ajeya_cotra I just stumbled across your account and find your statement interesting! "AI can get very powerful very soon." I would love to learn more about that. Is there any paper that you wrote that I can dive into? thanks so much 🙏
@ajeya_cotra The policy wonks are correct.
@ajeya_cotra I think often people have some first impression with technology and believe that they have a solid grasp of how it works and they proceed to have opinions (eg. chatgpt has no common sense).
@ajeya_cotra Please name the model that doesn’t confidently answer questions with nonsensical answers.
@ajeya_cotra Tested the “walking to the moon” example. Result was exactly what you’d want
@ajeya_cotra Sick of all the "mini tours". You need to establish an actual presence there, half-in-half-out is a house of cards and you won't be able to see it until it's too late.