"Training" an AI Agent for ONE Specific TASK with OpenAI-o1 API

[@apache937]

I think you should try finetuning 4o to becoming a redditor.

[@micbab-vg2mu]

thanks - ) now o1 API is available for Tier 3 users - I may play with it:)

[@clamr6122]

sweet. how have i missed this?

[@gurudaki]

Great video!Can you share the full code please?

[@KieranShort]

Perhaps all the rules are what has made the answers luke warm. What AI models do "fight back" and challenge?

[@Eysc]

whats the name of that code editor (I use vs code)

[@jana171]

The constant "pleasing" behavior is frustrating. I find that instead of pointing out idiotic code on my part, I’m often led on wild goose chases with suboptimal or incorrect solutions, just to avoid disagreement. It feels like the AI prioritizes completing prompts and using more tokens over accuracy, so I have to frame prompts as questions or doubts to get better results.

[@johleonhardt5637]

Amazing video been showing it to my entire organisation - do you think you could share the code for this?

[@emmanuelgoldstein3682]

Kris is the world's 2nd most intelligent Windows user

[@samyio4256]

I claim that extreme semantic repitition tailors the model towards hallucinations or certain biases. From experiments i can tell that repeating a token/word the llm produces significantly worse output compared to clean non repeating prompts.