ShareGPT: Share your ChatGPT conversations with one click
1–10 of 42 posts
Re: ShareGPT: Share your ChatGPT conversations with one click
#2I'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.
Re: ShareGPT: Share your ChatGPT conversations with one click
#3This project might actually seriously poison any future datasets for AI training. Some conversations are really fun though
Re: ShareGPT: Share your ChatGPT conversations with one click
#4How much of GPT generated text is going back right onto training new LLMs.
This has to eventually overwhelm organic human generated content doesn't it?
What's the way out of this
Re: ShareGPT: Share your ChatGPT conversations with one click
#5to 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."
Re: ShareGPT: Share your ChatGPT conversations with one click
#6How much of GPT generated text is going back right onto training new LLMs. This has to eventually overwhelm organic human generated content doesn't it? What's the way out of this
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.
Re: ShareGPT: Share your ChatGPT conversations with one click
#7I'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.
Re: ShareGPT: Share your ChatGPT conversations with one click
#8How much of GPT generated text is going back right onto training new LLMs. This has to eventually overwhelm organic human generated content doesn't it? What's the way out of this
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.
On the technical front there's a good paper here: https://arxiv.org/pdf/2301.10226.pdf, and a nice very approachable video explaining it here: https://www.youtube.com/watch?v=XZJc1p6RE78.
Re: ShareGPT: Share your ChatGPT conversations with one click
#9How much of GPT generated text is going back right onto training new LLMs. This has to eventually overwhelm organic human generated content doesn't it? What's the way out of this
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.
Correct output will be desirable. If you feed nonsense either human or AI generated you might break it.
Re: ShareGPT: Share your ChatGPT conversations with one click
#10How much of GPT generated text is going back right onto training new LLMs. This has to eventually overwhelm organic human generated content doesn't it? What's the way out of this
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.