|
I rarely use HN, so I apologise if my etiquette is off, but I think I've found 'a thing' and I'm really not sure where I can have a smort conversation with smort people about it, and well, this seems like a good starting place. The basic premise is this: After a few months of ChatGPT saving chats with its new Memory feature, I feel like it's starting to put me into context bubbles that I can't break out of. Example 1 - In one chat I asked it for some low carb recipes, using a particular ingredient, now in all chats where I ask for a recipe, the default suggestion is low carb using that ingredient. Example 2 - I spent some time asking it to generate images in Style X, and now I want to generate in Style Y, but it keeps giving me results in Style X, with artefacts of my Style X requests. (which I suppose may relate more to whatever underlying context is stored through the Dall-E API calls than anything) On the surface, this seems really handy, because it's learning my preferences, but I feel like it's starting to limit what the system is capable of generating for me, particularly when I give it new scenarios that I don't want to be influenced by previous conversations. And it's doing it at the speed of AI. Now, I know I can turn off Memory, or delete particular things/chats from memory, but I don't want the feature gone altogether, I just want to be able to say "for this chat, ignore all context from previous chats", but that instruction doesn't seem to work. Now, I might be missing something obvious, so I'm just wondering if people have found a way to effectively deal with this, or whether I'm just seeing some early signs that we're about to see a storm where ChatGPT is suddenly throwing everyone into very narrow context bubbles? |