Chase Lochmiller

speaker

453 appearances 4 recordings 1 series first heard May 2025 last heard 17 Sep

Chase Lochmiller’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 — 2 in all, peaking in Sep 2026 with 1.

Appearances

newest first · ▶ plays the moment

TBPN · Elon Musk Is Leaving Washington, Meta Partners With Anduril, Kim Jong Un's "Side Launch" Goes Sideways | Peter Rahal, John Andrew Entwistle, Ashlee Vance, Chase Lochmiller, Matan Grinberg, Jonny Dyer, David Hsu, Sushanth Raman, Rob Toews · 29 May 2025 podcast

Mm-hmm.

And then, you know, we kind of, you know, with the with the launch of Chat GPT, it really sort of catalyzed just massive investment and attention to purpose-built GPU infrastructure.

And we had done that from the ground up, all the way from energy, data centers, um, as well as uh uh managed infrastructure as a service at the software layer.

Yeah, I I I think just like the way that got spun up in the media was like so uh mystery.

like very early on.

Like I think anybody everybody was like, wait, like there's no chance this was like a couple of hobbyists that had like a couple of GPUs in their garage and they trained this model off of like that's just like not what happened.

They they did this

They did this training.

totally.

Um yeah, it was a couple couple guys with pens and paper just uh

I think it does affect some of the ways that we build data centers.

Um, but to the question of slowing demand, I mean the conversations that I'm involved in, there's like if anything, we're seeing demand accelerate.

Um and for you know, bigger, you know, bigger, you know, larger scale clusters.

Uh and uh

And and and just the overall demand is is increasing quite a bit for for for inference as well.

Um and I think I think you see kind of this uh

uh transition where, you know, folks will use a very large AI factory for a training run that, you know, gets some state of the art model.

Um, you know, that infrastructure is still useful for, you know, a long period of time uh to serve uh, you know, both inference workloads as well as you know any of these like post-training, uh test time compute scaling, you know, chain of thought reasoning models um that you know are are really basically uh

you know, taking inference queries and then thinking about them and sort of playing out a whole bunch of different scenarios, um, and then coming up with better, smarter, uh, more intelligent answers.

Showing 341–360 of 453 · page 18 of 23