Elan Barenholtz
speaker
2,007 appearances
4 recordings
1 series
first heard Feb 2025
last heard Jun 2025
Elan Barenholtz’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsNo recordings in the last 12 months.Older appearances are listed below; set an alert to hear about the next one.
Appearances
But you're in really engaging dis very distinct mechanisms that have some way of bridging.
Which is it's it's a non-answer.
I'm not gonna pretend that that is uh you know even halfway to a solution, but I do think it's a sketch of what how the how the cognitive architecture ultimately really is built.
Uh I I think that we we've now nailed down one piece, the linguistic piece, and we're able to say this is how it.
lives and this is how it operates and and it is and is and is autonomous.
And then we don't really have we don't have something similar for for perceptual space and for motor space.
We don't have something comparable.
We haven't been able to capture it uh successfully.
Maybe in robotics it's that's happening.
I don't know.
Um but we that's that's the kind of thing that the work that needs to get done.
to like different different language models just chatting with one another?
Great question.
Great question.
And and and and I and people uh
This is this is sort of one of the key challenges, um, you know, saying of of of of the sort of the identity hypothesis, uh, that we're doing the same thing, um uh which is there's sort of continuous learning.
And so um so there are two things that happen in large language models that we can call learning.
And one is the uh the actual shaping of the space, uh, which is really just uh determining, you know, the connectivity between neurons.
Again, you can think of it as a graph, you know, uh or you could think of it as as sort of just uh an an an embedding of determining the embedding.
Um but whatever it is, that happens during the course of training.
Showing 721–740 of 2,007 · page 37 of 101
← Previous
Next →