Kevin Russell
speaker
741 appearances
1 recordings
1 series
first heard Jun 2026
last heard 29 Jun
Kevin Russell’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 Jun 2026 with 1.
Appearances
There's only so much data that you can feed and then it has a negligeive neg uh negligent uh training.
It'll actually so as you're training the model, you'll see it drop.
And like there's all these different parameters that you say, oh now you're overfitting or you're underfitting.
You have to find that right sweet spot.
And that sweet spot is that conceptology.
Oh, I got it.
And once the model gets it, you don't have to do it anymore.
And once the bottle gets it, you don't have to do that.
That's like why I think this whole geometric interpretability will change the field as far as how much data do we actually need.
Do you need to take the entire internet?
Probably not because I've been able to do that.
Probably not, because I've been able to show on a 200 billion parameter model, I can take that conceptology.
And I thought it
with higher resolution and then you could put it on a small model.
And it's like, no, it's got a little bit more resolution, but there's only so much I mean, you've been a teacher, there's only so much you can teach before they get it.
And once they get it, then they learn the rest.
And that's just a very small uh like uh like the conceptology itself is very small relative to the data that it took to ingest it.
And I think that there's a happy medium that you just don't need it anymore.
I think that we did that and now we won't do that in the future.
So in this span of taking from a frontier two hundred plus billion parameter model that requires huge mainframes to run that model, um you still there, by the way?
Showing 381–400 of 741 · page 20 of 38
← Previous
Next →