Home
Creators
Search Recent Missing Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register Favorites Keys Review DMs Logout
Community
Telegram ThePornDude
Home Creators Posts Import Register Favorites Logout

Recording of today's seminar on GPT-4 (Patreon)

Published:
2023-03-25 20:26:38
Imported:
2026-06
haven't archived this post yet. have a subscription? use the importer!

Content

Some topics of discussion:

  • things we noticed in the paper which seem under-discussed
  • attempted tea-leaf reading regarding model size, compute, scaling laws
  • implications for software design
  • what are AI systems, as a medium or a material?
  • moral challenges for accelerationism

Chat transcript, for URLs referenced in the discussion.

Please don't share this video or the chat transcript publicly.

Files

Previews only

Seminar: GPT-4 (2023-03-25 09:04 GMT-7)

Comments

Anonymous

Thanks for sharing it!! Was on a hike this morning and missed it. Watching it now

Jorge Arango

Thanks for hosting this session, Andy.

Can Sar

I saw Stallman speak in person in the early 2000s (2003?) and he refused to give his talk because the recording codec was not open source (until they agreed not to record). Someone tried to make a "don't the upsides of spreading your message outweigh the downsides of the codec?" argument with him. It did not go well.

Can Sar

Do you personally pay for ChatGPT Plus or use ChatGPT actively? My sense is that even using the system at all is negative and paying is worse, but it is really quite useful already and a bit hard to stay away from.

Andy Matuschak
>>105034720

On using ChatGPT: my current (tentative) stance is that (quietly) *using* these systems, particularly for free, induces no meaningful marginal acceleration. So yes, I use them, for now. On paying for ChatGPT: I'm not very worried about the acceleration induced by one marginal subscription, so I'm not opposed to this now, though I haven't done it yet. It's somewhat (but not very) possible that I'll conclude that I'll want to avoid this as part of some kind of ideological policy, some sort of "moral purity", rather than because I think it'll actually induce any real acceleration.

Can Sar
>>105034720

Oh yes, I'm going off your comment about deontology vs consequentialism. I definitely agree that the actual impact is tiny.