ChatGPT with Rob Miles - Computerphile

Channel:Computerphile · Length:36:01 · Published:2023-02-01 · Views:517588

Watch on YouTube

Summary

Computerphile interviews AI safety researcher Rob Miles about ChatGPT: how it was trained with human feedback, why it tends to produce confident nonsense, and where it may or may not be dangerous.

Why HN recommends it

HN commenters send it to people confused about ChatGPT's quirks, explaining how human-feedback training makes it a pleasing yes-man that bullshits.

This video[1] cleared up a lot of questions for me as to the issues I've seen with ChatGPT. — HN, 2023

LLMs & generative AI · AI safety, AGI & impact

Subtitles (excerpt)

okay so you remember a while ago when we started talking about language models I just wanna I kind of just want to claim some points basically be like hey remember years ago when I was like I think language models are a really big deal and I think that like what happens when we scale them up more is pretty interesting but alignment is very important um seems to be what's being played out in the sense that uh chat EBT is very impressive but it's not actually like I don't think it's larger than gpd3 in terms of like parameter count um I was going to ask that very very question because you know we went from gpt2 and then we went oh gpt3 and it would seem like we were scaling up and up and up but actually has it just been smarter this time yeah well there's a sense in which it's better aligned that's one way you could frame it anyway because the original gpg3 was a language Model A pure language model um and so it in principle could do all kinds of things but in order to get it to do the specific thing you wanted it to do you had to be a bit clever about it like I think we talked about um putting tldr in front of things to figure out how to get it to do summarization this kind of thing there's a sense in which it's a lot more capable than it lets on

On Hacker News

I think it is still a problem that many people do not know how GPT really works. There is no part in GPT that can figure out something. So no intelligence and no learning of things. A good example is when you look at GPT writing poems. All poems that it writes are of the same structure and they are not very good. So it seems it does not know the difference between a sonnet and a limmerick. But wh… — fxj, 2023-04
This video[1] cleared up a lot of questions for me as to the issues I've seen with ChatGPT. My current operating theory goes thusly: Think back to "This Person does not Exist" [2], a site which generates a simulated human face. This works by randomly picking a vector into the latent space of human faces from a trained model, and showing it to the user. When you're using ChatGPT, you're getting a … — mikewarot, 2023-02
It might help quite a bit to watch this overview of how ChatGPT works, and especially how it was trained .[1] Human attention was too expensive for the training, so they made another AI to simulate the trainers. Unfortunately, the trainer simulator training didn't have an expressive enough set of options for the actual humans, which kept a lot of quite valuable feedback out of the whole system by… — mikewarot, 2023-02
It depends on your definition of cheating, but this is definitely not "using humans to answer all questions one could ask". Rather it's a way to tune language models to be more like assistants rather than "most likely continuation machines". For context I recommend watching the discussion on what chatgpt is/does[0], and what the open assistant project's aim is[1]. [0]: [VIDEO] [1]: [LINK] — f_devd, 2023-02
I’ve read that before, it is fascinating. GPT3 was just trained on source text, but ChatGPT has undergone an an additional very rigorous programme of intensive training bootstrapped by human feedback. This has evolved it in order to improve its performance as an information assistant, especially on specific tasks that have proved problematic in the past. This video explains the approach and some … — simonh, 2023-02

Related videos