Comment on We have to stop ignoring AI’s hallucination problem
muntedcrocodile@lemm.ee 1 month ago
Ais real power its ability to use tools and understand context form existing tools. For a Foss tool that uses an llm to do web searches and generate accurate(not guaranteed) results try my tool github.com/muntedcrocodile/Sydney
DdCno1@kbin.social 1 month ago
Are you seriously trying to push your ChatGPT "tool" in response to an article about language models like this one having substantial issues? "Not guaranteed" - yes, obviously, that's the point of the article - and from a quick look at your code, I don't see how this nonsense addresses any of that.
muntedcrocodile@lemm.ee 1 month ago
Its not chatgpt that’s just default config u can use the API endpoint to point to any chatgpt api compatible llm. Its can utilise ddg to search for web results then gives u an answer based on that. Most importantly it shows u the full log and u get to read it as it happens like bingAI but transparent so checking it is right there. I’ll add some screenshots to the readme.
DdCno1@kbin.social 1 month ago
Since the issue with hallucinations is shared by all LLMs, not just ChatGPT, this doesn't change anything.
muntedcrocodile@lemm.ee 1 month ago
Its transparent in its operation allows u to see what its thinking and catch errors and use ur own fine tuning that isn’t censored.