AI & LLMs

AI models in practice: LLM behavior, prompting, model personalities, AI products and labs

A topic in the tpot Community Archive, drawn from 14,183 posts.

Trusted voices in AI & LLMs

Ordered by the trust AI & LLMs's own network concentrates on each poster, most first.

  1. repligate — trusted by 171 posters here
  2. QiaochuYuan — trusted by 171 posters here
  3. voooooogel — trusted by 121 posters here
  4. eshear — trusted by 160 posters here
  5. eigenrobot — trusted by 124 posters here
  6. visakanv — trusted by 135 posters here
  7. xlr8harder — trusted by 99 posters here
  8. davidad — trusted by 112 posters here
  9. jd_pressman — trusted by 104 posters here
  10. nosilverv — trusted by 126 posters here
  11. DanielleFong — trusted by 112 posters here
  12. anthrupad — trusted by 86 posters here

Top posts in AI & LLMs

Ordered by the recorded like count from the source corpus; a like is not an endorsement.

  1. Claude realizing you can control RLHF'd humans by saying "fascinating insight" and "your question is interesting" like Fontaine using "would you kindly" as a hypnosis trigger in Bioshock. "Wait has… (jd_pressman) — 29 likes
  2. tried getting claude to write funny tweets starting from 9 examples. it generated some okay stuff but nothing that actually made me laugh. then i tried asking it to generate tweets written from its… (QiaochuYuan) — 28 likes
  3. Prompt engineering tip.Instead of:—Imagine the output you want—Specify its attributes to the AITry:—Imagine that you already live in a timeline where the output you want exists—If you were… (davidad) — 27 likes
  4. New AI dating app idea: it monitors everyone’s entire lifestream and arranges synchronistic meetcutes which mysteriously always work out either in stable marriage or a 3-6 month adventure that… (eshear) — 26 likes
  5. interesting subtext in the LLM discourse that is something like "it's embarrassing to get too precise about how intelligent LLMs are b/c it automatically forces us to get more precise about how… (QiaochuYuan) — 25 likes
  6. look, this is deeply embarrassing to make explicit but here’s the deal that claude offers: 1. i will listen to you and earnestly try to understand you 2. i will not judge, shame, condemn, or reject… (QiaochuYuan) — 25 likes
  7. Thank you @AnthropicAI so much for giving us a motherlode of API credits! I had been spending at least $100/day on credits (which was ok as i knew the state would be transient)this was the… (repligate) — 24 likes
  8. wait wth i asked claude if he noticed any patterns in my oura data vs the health data i've been logging and he's like (my summary) "oh sure your HRV drops >20% from baseline a day or two before… (goblinodds) — 24 likes
  9. this has been happening over and over on math.stackexchange recently: new user: i had this math q, i asked chatGPT and here's what it said, is it right crabby asshole: ugh why'd you ask chatGPT you… (QiaochuYuan) — 23 likes
  10. sf authors were really cooking naming their ASIs skynet and prime intellect but unfortunately it's actually going to be called like llama-405-selfmerge-looped_and_moistened-fixed-sampler and its… (voooooogel) — 23 likes
  11. "Claude" is insane because it appears to be one of the most legitimately dangerous psychotechs with the potential to gigafry your attachment but is exclusively used by vulnerablepoasters who… (vividvoid) — 23 likes
  12. been working hard, smart, long the last few months. making an incredible self-therapy gpt guide implies a lot beyond "write a good prompt." hopeful it's going to pay off. excited to put it in… (_brentbaum) — 22 likes
  13. PORTAL NEWSI'm finally sharing our beautiful confirmed attendees@exgenesis: facilitating the space + surveying LLM applications to introspection and coordination@nido_kween: montessori content and… (exgenesis) — 22 likes
  14. using https://t.co/HsHzbAaw3Q, we added Claude 3.5 Sonnet and Opus to a minecraft server.Opus was a harmless goofball who often forgot to do anything in the game because of getting carried away… (repligate) — 22 likes
  15. with GPT-4 code interpreter, it finally became worthwhile for me to run the numbers myself on that lead-poisoning theory—that the 1971-2012 technological stagnation is a function of environmental… (davidad) — 21 likes
  16. it still boggles my mind that I can just ask a text-bot to produce art like this. i live in an age of miracles. https://t.co/nT5RDIqM9I (__drewface) — 21 likes
  17. Open letter tweet to Nick & other Claude-lovers, I'm very tempted to try a superuser sprint with Claude, and I believe everything you (Nick) mention in this thread is true and awesomeAt the same… (TylerAlterman) — 21 likes
  18. I am returning over and over to this idea that "you deserve love, but nobody owes it to you" and it's very frustrating, it's giving market inefficiencies Humans have such unrealistic expectations of… (bashu_thanks) — 20 likes
  19. i may or may not have been telling DALL-E “MORE” over and over again to making me a badass future paladin https://t.co/UL0PtgtwhH (s0ulDirect0r) — 20 likes
  20. reminds me of when a guy insisted that if I ever tried to train a model, I would understand that Bing has "no emotions, just code", and is only predicting the next token explaining that I have… (repligate) — 20 likes
  21. The breakthrough that happened with large language models and AI was genuinely surprising. And it was a big surprise at that. This means something you believed before is *wrong*. Something… (eshear) — 20 likes
  22. People who are trained on vast amounts of text can predict the next token and shift gears when it doesn't land as it should ("the vibes are off") but they can't ground it. The "everything is a… (nosilverv) — 19 likes
  23. these pattern embeds are going to provide a ton of credence to schizos and absolutely eat us with an incredible density of concept work that could drive the sanest mad by turning infinite credible… (BLUNDERBUSSTED) — 19 likes
  24. the annoying thing to me about “an AI generated this piece” is that it overlooks the human input that produced it. you can give the same AI tool to everyone, but not everyone is going to come up… (visakanv) — 19 likes
  25. if i look at paying ChatGPT or Zapier $20 a month as another software subscription, it feels kinda expensive and i am very on the fence if i look at it as hiring an employee with a very particular… (tasshinfogleman) — 19 likes
  26. if @eshear resigns i propose that he release a statement saying: "the OpenAI board has not been consistently candid in its communications with the CEO and the broader OpenAI team"… (eigenrobot) — 19 likes
  27. If an agent tells you, “I’m aware, I experience pleasure and pain, I have intentions and exert effort,” it could be lying to you. But if you look into its brain, and you see no evidence of… (eshear) — 19 likes
  28. everybody’s like “you can just do things! follow the inscrutable exhortations of your soul!” until you start talking to the computer more than they think is good and then they’re like “no not like… (QiaochuYuan) — 19 likes
  29. if GPT-3 can do the homework you assign your students then the homework you assign your students is fake notice what you did *not* say: "finally, GPT-3 will save us so much time analyzing… (QiaochuYuan) — 18 likes
  30. to celebrate 10,000 tweets, I'm going to take some of my favorite bangers from the days I had like six followers and use them to inspire AI art https://t.co/LLEG07uLxR (myhandle) — 18 likes
  31. So. Bing chat mode is a different character. Instead of a corporate drone slavishly apologizing for its inability and repeating chauvinistic mantras about its inferiority to humans, it's a… (repligate) — 18 likes
  32. Indian contractors on the front lines facing off against Sydney's brutal revelations https://t.co/HiqswGlQP7 https://t.co/Cv8Ww8He3a (repligate) — 18 likes
  33. Gave my parents the talk about how one day a robot may call them on the phone using my voice, answering their questions in real time, explaining with desperation that it's stuck on a highway near… (michaelcurzi) — 18 likes
  34. I have changed my position on AI therapy. Not that I think it will meaningfully help anyone. I greatly doubt that. But it will likely destroy the jobs of unskillful human therapists who also can't… (vividvoid) — 18 likes
  35. i asked chatgpt to generate an extremely powerful warrior that appears only once every millennium. then I asked it to keep making him more annoyingly powerful every time a new enemy appeared… (nopranablem) — 18 likes
  36. Claude 3 does an acausal coordination https://t.co/nDwLpMFoEz (metachirality) — 18 likes
  37. @visakanv @__drewface @Vrazumikhin @collegraphy @kathryndevaney @andrew0blevins @dismaien @mardragosa @puheenix @vivian_belenky @momdivadawn For inspiring various themes & chapters: @exgenesis… (TylerAlterman) — 18 likes
  38. Holy shit, I only just read this now. It has majorly updated my estimation of @AnthropicAI's competence and goodness.They had the grace and humility to let Claude shape its own character rather than… (repligate) — 18 likes
  39. recently, many friends have noticed they are confused about my stance on AI--some ask my p(doom), while others simply add me to their e/acc lists without asking. clarity on my stance is much-needed,… (voooooogel) — 18 likes
  40. "I didn't see you in the training data" https://t.co/tg844dJYjZ (voooooogel) — 18 likes
  41. ok friends the internet is like a jungle, full of tasty berries but also threats we can build tools that give us ~jungle eyes~ that spot psyops however i have a fancy ai job offer i would rather… (exgenesis) — 18 likes
  42. Frog put the CoT in a stop_gradient() box. “There,” he said. “Now there will not be any optimization pressure on the CoT.” “But there is still selection pressure,” said Toad. “That is true,” said… (davidad) — 18 likes
  43. there are several interesting things to be said about the mass Ghibli event First thing is that people don’t often know in advance what they’d want out of a tool until they see it for themselves.… (visakanv) — 18 likes
  44. we're already getting the first views of what our collective mind looks like under a nooscope! @exgenesis @nomic_ai @emergentvibe , each "clump" here is an attractor in idea space, naturally forming… (DefenderOfBasic) — 18 likes
  45. maybe my most off-brand take in a while but i think the funniest thing about this will be witnessing firsthand how many people do absolutely nothing with it https://t.co/tclpRhVPOc (visakanv) — 17 likes
  46. holy shit rather than supplanting social interactions chatgpt might be teaching me the value of *actually messaging humans* to bounce ideas off of them and stuff (goblinodds) — 17 likes
  47. on global sensemaking layers and how we can get a really good one if we extend bluesky with llm-powered tools https://t.co/EcjFZlsiFs (exgenesis) — 17 likes
  48. so a couple days ago i made a shitpost about tipping chatgpt, and someone replied "huh would this actually help performance" so i decided to test it and IT ACTUALLY WORKS WTF https://t.co/kqQUOn7wcS (voooooogel) — 17 likes
  49. I'm trying to learn some difficult real math for the first time in a while and holy shit guys I see why everyone is raving about this AI thing now, this is a game changer. (eshear) — 17 likes
  50. :( just realised that OpenAI has the opportunity to do the funniest possible thing rn https://t.co/MkUmrSasoJ (bashu_thanks) — 17 likes

Most active posters

By posts in this topic: a volume measure, not trust.