Md-Mq

Engineer, startups, philosophy, music, coffee ...
Building an AI control plane, evals, gpu infra/scheduling ...
Alum @ BNP, EIB, Kayak, Dubsmash/Reddit, YC ...

[Md-Mq @md__mq]

@ylecun @OmarSAlolama I generally agree with Yann, premature/over -regulation is harmful to science as well as to the startups ecosystem.
But in this example the technology was "banned" not "regulated".

[Md-Mq @md__mq]

@transitive_bs @OfficialLoganK @OpenAI The API will definitely evolve when users start using the assistant in prod.
I already see how a UI can enable users to @ multiple assistants similar to how we interact in slack, and @ another assistant to analyze the result.

[Md-Mq @md__mq]

I agree but there are use-case where also you would want that, here are 2 use-case:
* A/B testing 2 or more assistants to compare their performance on a single thread, in this case you might want to continue the discussion based on the best result.
* Map-Reduce: You would to invoke 2 or more assitants and perform some ensembling, comparison, ..., on all results.
I think the best is for the list API to expose a filter by assistant, but not sure how intuitive.

[Md-Mq @md__mq]

Threads can be executed by multiple assistants that's why they do not have explicit relation. It's a many-to-many with the run as an intermediary table.
You can for example have 2 assistants, create a thread, and decide to start two runs based on that thread with each of the assistants.