Buttons not working properly for OpenAI models.
Hi, I have been discussing this issue for a couple of days now but I got no feedback.
I have been experimenting with Agents as they are a convenient way to create assistants. To be able to harness Carousels and Buttons, the only consistent and reliable way is to use Anthropic models. However, these models are extremely pricey, and the subscription is not built to withstand high traffic with such models.
As soon as I switch to cheaper models (e.g GPT 3.1 mini), the behaviour becomes highly inconsistent, and the agent immediately loses it's quality.
Here's an example to reproduce :
You can try it yourself, using OpenAI models vs. Claude, even 3.5 Haiku will do a better job than GPT 4o.
This seems like an API problem internally rather than a prompting problem. Please help
0 Replies