David Cox

speaker

634 appearances 1 recordings 1 series first heard Jan 2026 last heard 13 Jan

David Cox’s voice in public audio — every appearance, attributed to the second.

Trend

recordings per month · last 12 months

Recordings per month over the last 12 months — 1 in all, peaking in Jan 2026 with 1.

Appearances

newest first · ▶ plays the moment

The Neuron: AI Explained · Why IBM Wants AI to Be Boring · 13 Jan 2026 podcast

It doesn't force the model to do something that it wouldn't naturally do.

And that sounds like I'm almost like it's coercive or something, or it's like you're going to make the model mad.

But the reality is, it's more like going with the flow.

It's like if the current wants to take you this place, and all we need to do is paddle a little bit to go that way,

That's the way to go versus saying I'm going to like dig a trench with a backhoe, like straight line across, which is what some of the other kinds of training methods end up doing.

So it's very, very powerful.

It's very, very effective.

And it's amazing how quickly things are shifting.

It is.

Yeah, there's a crazy number of words for small.

So we have a small, which is kind of around 30 billion, 32 billion parameters, 35 billion parameters.

I can't remember the exact number.

And then we go tiny, then we have a micro, we have nanos.

We actually wanted to release a mix of different models.

So for most of the versions we have,

dense ones.

And we have, we have regular transformer versions like vanilla.

And then we have these hybrid ones we talked about earlier, which I guess are like, I don't know, pistachio cream or something out of the ordinary ones.

And then there's these faster ones.

And part of the reason we're doing that we're using multiple versions is

Showing 381–400 of 634 · page 20 of 32