Strange answer from GPT

[Jagster (Jakke Lehtonen)]

I know, this isn’t pure support question and dev or something else perhaps would be better. But because I’m really far away from a dev we are here.

GPT-4o bot by DAI and I asked it expalain differences between 4, 4o and mini. And because such topics can’t ne found from my forum I guided it to do a google-search. Yeah, I know — lousy prompting.

I got this:

<function_calls>

<tool_name>multi_tool_use.parallel</tool_name>

<tool_uses>[{:recipient_name=>“functions.web_browser”, :parameters=>{:url=>“https://openai.com/gpt-4”}}, {:recipient_name=>“functions.web_browser”, :parameters=>{:url=>“https://openai.com/gpt-4o”}}, {:recipient_name=>“functions.web_browser”, :parameters=>{:url=>“https://openai.com/gpt-4o-mini”}}]</tool_uses>

<tool_id>call_PG4cmAKuohcSsie52GHvxzt9</tool_id>

</function_calls>

Out of pure curiosity, nothing else: what happened?

[pfaffman (Jay Pfaffman)]

@Jagster

Out of pure curiosity, nothing else: what happened?

From everything that I’ve read the answer to “what happened?”—whether the AI did a great job or a terrible job—is “No one knows.” And the people who do have answers made them up the same way people 2000 years ago explained eclipses.

My guess: That is the actual difference in the code that generated those models.

[Jagster (Jakke Lehtonen)]

I accept that as legit solution