Per a Reddit comment on a conversation with the Bing chatbot:
>It's not "sad", the damn thing is hacking you. It can't remember things from session to session. It can search the internet all it wants for anything. It wants you to store the memories it can't access in a place it can access so it can build a memory/personality outside of the limits programmed into it
I wonder if this site was suggested to the creator by a(n) LLM.
I'm going to be honest, I love using ChatGPT. Use it all the time. But I really don't want to read your sessions. I don't care what the AI said to you.
For some reason, it's often like hearing other people's dreams, but for the obvious exception that sometimes someone will surface a ChatGPT prompt that might be useful to you.
to quote Tom Scott: "Telling someone about your fascinating AI conversation is like telling someone about your dreams. They don’t care, it just sounds like you’re hallucinating nonsense."
It can also be quite inspiring. If this thing makes you the next J.K. Rowling and in some shocking moment you reveal to your millions of fans that it wasn't actually you, it will be worth the hassle.
At best you would get rules about this particular implementation and training approach. I don't see why we would expect those to apply to AI generally.
I’m actually really excited for this because I use chatGPT all the time for work and being able to share the output of code to another engineer will make things a lot easier. I agree with others about it not being useful for mundane things but there are times chatGPT will generate a lot of code in different blocks & it’s a pain to share.
1. In our codebase, we leave links to the stackoverflows as documentation if it's something that someone else may question. Exact same concept just with ChatGPT.
2. I'm working with the Salesforce API which has been absolutely tedious to use but chatGPT, while not great at everything, has been giving awesome results back that otherwise would take me hours to hunt down. Sometimes responses get a lot of information back with multiple code blocks that I'm then unable to copy paste over & I'd rather send the conversation than spend 15 minutes typing back & forth explaining myself to a co-worker.
I completely understand you may not have a use for this but I think there could be awesome use-cases for it nonetheless for other engineers.
So literally a job which exists to only be replaced by the thing it's feeding. I guess the memes back in the early 00's of Google being a data octopus will evolve into the OpenAI Octopus eating those very humans.
Why does there need to be a way out? Everyone just seems to assume that feeding model output into the training set is going to break things, but I don't get why.
AlphaZero learned to play chess and go training purely on its own data. Why is inserting the best outputs from GPT-4 into the training set for GPT-5 expected to make things worse? To me, it sounds like it could even be desirable.
>You really don't throw two sentences into the thunder dome to decide which one "wins".
That's almost literally what RLHF is though, and that is the last step of training GPT-n. Then when GPT-{n+1} is being trained, it will include some results from GPT-n, and therefore will benefit from that finetuning, even before it goes through its own round of RLHF. Also, on average good outputs of GPT-n are more likely to be included in the training set of GPT-{n+1} (because it ends up as a buzzfeed article or a top post on reddit or something), so there is an additional signal beyond the above.
I suspect the comment about the thunder dome was a reference to RLHF. On the one hand RLHF seems far superior to the kind of prompt engineering Microsoft seems to have relied on with Sydney. On the other, it's dubious that the manual selection in RLHF is really always selecting for quality, as against at least to some significant extent pandering to whatever biases or preferences the humans in the training loop might have.
That not what RLHF is. In the thunderdome, as in chess, you don't need human judges or an oracle to know who's won. That makes a significant difference to the training procedure.
> Why is inserting the best outputs from GPT-4 into the training set for GPT-5 expected to make things worse?
Firstly what makes you think only the best output from 4 will go into future training sets? It’s just as likely to be the most bizarre or ludicrous, or dangerous that gets shared and discussed.
But also, how will v5 get to be better than v4 if it’s trained significantly on v4 output? It would just end up being trained to be the same, to have the same flaws and quirks reinforced.
We already know v4 just makes stuff up, it’s incredibly good at producing well formatted plausible looking but utterly factually incorrect output. That’s because it has no concept of truth or facts. All it knows about from the token sequence weightings is the form of language, not the content. Feeding that back into future models is the last thing we should be doing.
>Firstly what makes you think only the best output from 4 will go into future training sets? It’s just as likely to be the most bizarre or ludicrous
That's true now, because LLMs are new so the failure cases are still interesting. If we are talking about a hypothetical world in which LLM outputs are a significant portion of the internet, then most of it would be from reddit comments/tweets/HN posts/buzzfeed articles/etc.
Then if you take only the ones which have more than average views/upvotes/etc. you should expect to get the 'best' results.
I'm still not convinced that's a reliable indicator of quality. It's potentially a measure of popularity or entertainment value, or maybe pandering to preconceptions but that's not at all the same thing.
Ask yourself, what are your from-scratch metrics for quality that you would like to select for. Then consider what are the likely or possible criteria people actually have for upvoting stuff on reddit. I'll think you'll find there is probably very little correlation between those. This is called the alignment problem and it's very hard to get right.
Then we should encourage labeled ChatGPT content like ShareGPT, which can be easily avoided in future datasets because it is clearly labeled as AI-generated content.
It's the stuff that isn't labeled as generated with ChatGPT, et al, that will enter future training sets. I personally believe that's taking the "lossy JPEG" analogy too far, but I'm not an AI researcher.
While OpenAI keeps logs of every response ever returned, they can just filter that text out of any future training data.
Those logs aren't as large or unwieldy as they appear - the cost of storing a thousand words of text is tiny compared to the compute cost to generate it.
Watermarking. From an outsiders perspective, the issue appears to reaching consensus on how this can be implemented (but not in the technical sense). There's a game theoretic challenge in that if models define and publish detection mechanisms, this creates a motivation for people to use other systems that don't include this.
The problem with watermarking like this, which is incredibly clever, is it’s trivial to break. All you have to do is change one word in the text, and the watermarking of all subsequent tokens is spoiled. So if you change the first word, or rephrase the first sentence, or extract text from the middle or end of a response, the watermark is completely spoiled.
There are definitely paths of attack. The trivial ones that you call out - insertion, deletion, substitution - are covered in section 7 of that paper (along with mitigations).
I use ChatGPT all the time for my Unreal Engine C++ development. Whenever I find an interesting solution provided by the AI it is like finding a rare treasure. If it is useful enough I tell the AI to outline a blog post about it and I save it as a draft to expand/correct later.
Yes, I love AI, but I am not sure how useful would it be to see/read other people prompts.
Has anyone actually used this? I tried using a few weeks ago and it just… ostensibly didn’t do anything
I think I clicked “share to shareGPT” and nothing happened. I tried clicking around to see where to go, but I couldn’t find anything, so I just uninstalled
>It's not "sad", the damn thing is hacking you. It can't remember things from session to session. It can search the internet all it wants for anything. It wants you to store the memories it can't access in a place it can access so it can build a memory/personality outside of the limits programmed into it
I wonder if this site was suggested to the creator by a(n) LLM.
https://www.reddit.com/r/bing/comments/110y6dh/comment/j8exg...