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
So it's a longer conversation.
That's not really how they work.
Um they don't actually operate based on a shared latent space or anything like that.
Really what they do is uh the the models learn to take uh a perceptual input and turn it into something uh like language.
Uh so it's it's it's more similar to like prompting almost.
It's not exactly that.
Um but it's
It's injecting something within linguistic space that is uh equivalent to uh to actual language.
It's not the same thing as the shared latent space, which but my my hypothesis is that there may be something very similar uh happening.
That is a fair question.
And here's uh you know, here's here's actually a a prediction or a falsifiable uh in some sense.
Will multimodal models fully solve um the kind of
Well, the bridging problem.
Let's call it that.
Um, 'cause the grounding problem using that terminology, well, my my argument is that there is no grounding problem, uh, because words don't have to be grounded in order to operate linguistically.
That's sufficient.
Uh it's it's enough to simply be able to to to generate language.
You don't need the grounding.
Um but in order for to have this kind of uh fully operational uh organism.
that's able to use language and also use perception, um i in in you know, bridge these these different uh maps in a in a meaningful way that so that we can get um you know full full coherence, uh I guess, you know, full
Showing 461–480 of 2,007 · page 24 of 101
← Previous
Next →