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
That's really hard in a world where you're asking it to write you emails or help you with a PDF or help you how to express ideas or you're just using it as a thought partner.
And the the biggest realization that I feel like I've had
We we talk so much about human in the loop, quality, evals.
How do you have product builders?
How do we all read more transcripts or spend more time understanding what the requirements are?
And every AI product that's in front of users already has one human in the loop essentially doing quality control and evaluation.
And that's the the user, implicitly or explicitly in how they're steering it and how they're guiding it and what they're pushing back on, that they are providing signals about what is the AI doing well, where is it missing the mark?
Where does it not have the right context?
Um, and where I've been most successful in my own usage, and I feel like I've been able to save more of my own time, because you're you're right, like to get really good results out of AI, I often have to
look at the outputs a lot and iterate on them a ton of times.
The biggest gain that I've had is just looking, taking my own transcripts, taking my own prior interactions.
And you can do this on the product level too for your users as as kind of an aggregate.
And using mining that to basically come up with what are the right skill files, what are the right things to encode now, eventually closing the feedback loop.
You have all these users, you yourself in your sessions, you're revealing your preferences, you're showing how you like to steer, you're you're revealing your requirements.
The data, those learnings are already there.
How can we just encode them in a way that makes them systematically now where I think this gets a little tricky.
Is
The more that you have the AI running on the background on a really complicated task, the m the harder it gets to review its output because the amount of work that it's doing, the amount of thinking that it's even doing behind the scenes that you don't even get to see the tokens for, that is it is is exploding exponentially.
And I do really worry about a trend where we are using AI more and more from a delegative perspective.
We're having we're thinking, hey, I just need to write a really complicated spec and then I need to go have it run loose and then I'll come back 12 hours later and I'll see if it looks okay.
Showing 301–320 of 399 · page 16 of 20
← Previous
Next →