Rovo Agent temperature selection and resulting use case constraints

Are you considering to allow control over the temperature selection down the road (example)?

Our use cases require as deterministic as possible responses, so despite the ingrained non-deterministic aspects, the inability to tune the temperature towards a more deterministic result (or the temperature even changing under the hood) potentially subverts the value proposition.

2 Likes

Hi @sopel ,

We are aware that apps need to be able to rely on Rovo returning results consistently and that some apps will be more sensitive to variance than others.

I don’t think any particular solution has been identified, but I believe it will need to balance our ability to abstract Rovo away from any particular AI model, whilst also providing apps with some kind of control of model selection/parameters.

If we think of Rovo as an API, I don’t think the responses returned by this API have any particular guarantees so my suggestion would be to ensure your app is resilient to varying results from Rovo. For example, make sure there is transparency relating to any AI generated content and ensuring there is a human in the loop where data is created/mutated/deleted. Perhaps it is also possible to minimise the amount of pure logic that Rovo is required to perform by decomposing larger AI tasks into smaller AI tasks that are controlled by the app?

Regards,
Dugald

3 Likes