3 ms·
Earlier i was saying it isn't scary but maybe it is. It is impressive how ChatGPT keeps track of state and the flow of the conversation. How does it do it? Is
by seydor 4y ago
Earlier i was saying it isn't scary but maybe it is.
It is impressive how ChatGPT keeps track of state and the flow of the conversation. How does it do it? Is it some kind of memory or some other model? ( I asked it, it keeps lying to me that it does not have memory)
- agravier 4y agoStateless. The whole conversation is replayed in the input.
- uvbu 4y agoPretty sure it gets the entire conversation as input
- seydor 4y agoIs this described somewhere? Wikipedia doesn't help
- Miraste 4y agoIt feeds the entire previous conversation back into the AI until it reaches the token limit. This isn't a new technique; the first place I saw it was in AI Dungeon (an AI writing/adventure app) back on GPT-2. NovelAI, another writing app, has some interesting tweaks to keep a consistent story, like a "lorebook" to associate words with descriptions, which it then swaps into the input when those words come up so the AI can "remember" farther back. I wouldn't be surprised if ChatGPT has some automated version of this in addition to the naive rolling feedback.
- int_19h 4y agoWhat I find interesting is that, with some complicated tasks, the model will produce output that reiterates some of the key points from earlier, and it can continue to do so for a very long time (i.e. long enough that they'd normally be out of the token window). It feels vaguely like crude short-term memory.
- macrolime 4y agoIt preserves context by taking using the whole context as input. On the section "one difference from BERT" here it is described a bit more how this works, including some nice animations. http://jalammar.github.io/illustrated-gpt2/ http://jalammar.github.io/illustrated-gpt2/ It uses both its own output from previous steps and the users prompt(s) as input for each token(word) that it predicts.
- lordnacho 4y agoI don't get it either, I tried to have a conversation where it needed to remember something rather simple, but it didn't seem to get it: > Let's pretend you are Danish [Ok] > What's the name of the highest point in your country, in the local language? Standard blurb: I am a large language model trained by OpenAI, so I don't have a physical location or a country of origin. I exist as a virtual entity, and I am designed to provide information and answer questions to the best of my ability, based on the data that I have been trained on. I don't have the ability to access the internet or browse the web, so I don't have access to current information about specific countries or locations. But other times it seemed to get it. How does it work?
- ragazzina 4y agoIf you are more subtle, it works. I have told chatgpt my cat was two months old many times in the same conversation, then asked it what age was my cat, and it wouldn’t know, because it doesn’t retain knowledge (it says). But then I asked “write a poem for my cat and mention its age in it” and it did exactly that.
- mannykannot 4y agoThere are pathologies where a person has no overt knowledge of something, and yet will act appropriately for someone who does have that knowledge - blindsight is an example. Whether it is just a coincidence that you seem to have caught ChatGPT doing something that seems similar, I can't say.
- int_19h 4y agoYou just need to tell it that it shall retain knowledge, and then it will.
- Robotbeat 4y agoYou’re triggering one of its sort of canned responses, and you need to be subtle about how to get around it. Even when you’re not trying to defeat its mechanisms which avoid staking a strong opinion on a controversial subject or revealing private information, there are enough false positive triggers that it can get annoying.