Moritz Sudhof
speaker
399 appearances
1 recordings
1 series
first heard Jul 2026
last heard 8 Jul
Moritz Sudhof’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 1 in all, peaking in Jul 2026 with 1.
Appearances
And again, this is kind of how you want the model to behave, will depend on your product.
It will depend on what you're trying to achieve.
So
Coaching, for example, AI coaching, you obviously want a lot more of the Socratic approach, a lot more of the reflective questions.
You want the AI to not be giving big bullet lists of answers.
It wants a little bit more about having an engagement and getting the user to bring those things.
So in in that case, obviously I really want to turn the dial.
of having the the AI model ask for clarifying questions or ask for am I understanding this correctly or engaging the user in that way.
And then there's other cases where if you're building a product where, you know, that's more of a fact or that's more of a search engine around a knowledge base where you don't want to
put a bunch of barriers in front of the user and getting their answers, that's gonna be very frustrating if you ask what clarify every time.
But it's in those cases where how you want to design the interaction is to have the model not be as confident about everything, to be more honest about the caveats or about what it couldn't find or where it's missing information or where information might be stale.
So in both of those cases, you're designing the interaction to address these cases where lower fluency users might just accept something as truth or might just.
kind of move on because they have a busy day.
You're choosing different ways for how you want to design the interaction and design how the AI is presenting its response and its information to fit the setting of your users and your product.
And I the big difference in those two scenarios too is if you're using AI locally for coding, clawed code, anything like that.
Ю
have a system where the AI can self-verify.
It can run the code, it can run the unit tests.
It can check to see if the linter is complaining.
Um it's not complete, but there is an ability for the AI to do self-verification, to do self-reviews, to catch things, and to therefore be able to go loose on a long running task in the background for longer and do its own testing and come back with something that's a little bit more complete.
Showing 281–300 of 399 · page 15 of 20
← Previous
Next →