Is this a bug? the llm answered itself previously, so it said “To add to my previous description”, but the previous description was not shown to the user. can we allow the llm agents to show intermediate steps (make a dropdown like in chatgpt when it thinks) to the user for more clarity and a better user experience
1 Like
This is interesting… I haven’t actually had the LLM do this before.
There aren’t actually any intermediary steps that are hidden here, so I’m thinking the LLM misunderstood the intermediary steps that are shown (like “Searching the web for …”) which caused it to hallucinate and act like its response was adding on to what it already said.
Haha. Have seen ‘hallucinate’ used a few times over the last couple of weeks in multiple videos. That included ChatGPT, Claude and OpenAI.
I hope we’re not getting in the way of AI. Dumbing it down to a human level. I mean, next it will be day-dreaming of happier times and taking sick-days…!!!
1 Like