Hacker News new | past | comments | ask | show | jobs | submit login
Ask HN: Is there large language model specifically for hallucination?
4 points by thatgerhard 10 months ago | hide | past | favorite | 4 comments
So almost, answering questions without knowing any facts. If that makes it clearer.



Couldn't you just make that part of your prompt?

ChatGPT lets you pre-prime your chat sessions with an automagic prompt that basically initializes your session. That seems like the right place to prompt for answers that are akin to hallucinations.


Have you tried increasing the temperature on your model or are you more interested in the theory of such a system?


Ideally I would want to set up my own. Would this just be a "tweaked" version of a model and dataset on huggingface? Or just a setting that needs to change once I have one up and running?


In the case of what you're trying to do, you might be able to just increase the temperature a bit and instruct the model to lie to you, like this...

https://chat.openai.com/share/78d5c6a8-933b-4eae-bbc3-7737cd...




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: