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
We found that there is a very small percentage of users that are able to recover from these failures.
They are what we call high fluency users, and they behave very differently from the majority of AI users.
High fluency users, what they do is they will push back, they will be skeptical, they will ask for confidence assessments, they will offer a
additional constraints or ideas or play devil's advocate.
These are users who are highly augmented.
They're highly engaged and they're never treating the AI's output as a stone tablet.
They're always pushing back.
And you have the majority of other users that treat it a little bit more like a slot machine or a vending machine.
where they put in a prompt, if they don't what like what they get, they say try again or they walk away.
It's those kinds of users who have the highest share of invisible failures.
They're getting the most kind of quality problems back from the model, but it's not being surfaced or repaired.
Um and it's those high fluency users who turn those invisible failures into visible ones.
Like, hey, you got that wrong.
Like, what do you mean this is this isn't a green pepperoni isn't a vegetable?
Like, try that again.
They're the ones who are actually reviewing the outputs and then pushing back.
We did a causal analysis or or tried to to do actually rigorous modeling to see, well, do these behaviors actually matter?
Do they actually change what users get out of the AI?
And it's those expert users that 3% of highly engaged augmented users.
They encounter more failures.
Showing 241–260 of 399 · page 13 of 20
← Previous
Next →