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
Because now that we know that language is capable of doing the stuff that it does it, now that we know it has the properties to and I'm sort of giving away the the punchline to self-generate based on its
It's unavoidable that to think that we are using the same basic mechanism and principles because it would be extremely odd or uh to think that we have a completely different orthogonal uh method.
uh for generating language.
Or un put it differently, if we are using completely different mechanisms than the language models, uh then it's extremely unlikely that the language models would work as well as they do.
The fact that language has this property, has these this property that it can self-generate, the fact that that uh property actually leads to human level language to me forces the conclusion that there's only one way to do language.
Um and that one way is the same in
In humans and in machines.
Right.
And so that that's sort of the the critical question.
Is this mimicry?
Right?
Is it is what the the the models are doing, in a sense, learning a kind of roundabout uh technique that uh captures some of the superficial uh components of language in humans, but ultimately it's a completely different uh approach.
And so, you know, my argument is is really from uh the the fundamental simplicity of these models.
So
Let's just talk really quickly about uh how large language models work, things like ChatGPT.
What they are doing, uh, is learning that
Given uh a s a a a sequence.
Uh, you know, let's say the sequence is uh I pledge allegiance to the and then uh the model is being asked to do this thing called next token generation.
What's what's the probable next word?
And we're gonna we'll say word for the for the purpose of this conversation.
Showing 21–40 of 2,007 · page 2 of 101
← Previous
Next →