3 ms·
(I may have misunderstood your comment to some extent, but I'm going to send this reply anyway even if just to clarify for anyone else who might misunderstand.)
by Cyphase 4y ago
(I may have misunderstood your comment to some extent, but I'm going to send this reply anyway even if just to clarify for anyone else who might misunderstand.)
---
I agree with "be careful what you send to the chat bot", but let's clarify some things in case you or someone else reading your comment is misunderstanding.
LLMs aren't immature AI brains that "may even permanently ingest user data for training purposes". They're just models, which are represented by an architecture described in readable source code, and weights derived from training.
There is a very clear delineation between inference and training. Models are static when being used for inference. You don't need to "untrain" the model after you ask it something; you never trained it in the first place. Running inference does not change the trained weights.
If you're talking about OpenAI specifically saving ChatGPT data for later training purposes, they absolutely are doing that; they aren't hiding it. But that's a purposeful "let's take this data and use it for training", not "oh no, our immature tech accidentally ingested prompt data, how do we untrain it"?
- semiquaver 4y agofrom the “reliably isolate sessions” reference I have to assume they are referring to this: https://news.ycombinator.com/item?id=35291112 https://news.ycombinator.com/item?id=35291112
- Cyphase 4y agoI figured the same; I didn't address that point.
- fl7305 4y ago> Running inference does not change the trained weights. That's true today. I don't know how many days (or hours) away we are from GPT-4 running a LoRA-pass to update its weights after each round though?
- Cyphase 4y agoThat would be an explicit decision by OpenAI, not a result of immature tech.
- fl7305 4y agoSure. But my point was that it is not an inherent feature in LLMs that they are frozen in time. Fine-tuning the entire model is very expensive. But fine-tuning a tiny parallell piece using LoRA is cheap both in CPU cycles and storage. OpenAI could already have implemented an auto-update feature without telling us. In the future, I can see them selling a premium feature where you have your own LoRA-addon that gets constantly trained on your interactions with it, so you get your own personalized GPT-4.
- alach11 4y ago> If you're talking about OpenAI specifically saving ChatGPT data for later training purposes, they absolutely are doing that They claim they're not retaining data through the API.
- Cyphase 4y agoThat's correct (after 30 days). When I say ChatGPT I mean the web-based frontend product, not the models behind it which you can also access via API.