Aaron Levie
speaker
937 appearances
2 recordings
1 series
first heard Jul 2025
last heard 5 Sep
Aaron Levie’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 Sep 2026 with 1.
Appearances
No, you th that was w that was some of the best content ever produced.
Uh so I d I I don't know why they didn't charge for that one.
Yeah, I I you know, I first of all, I even even the open versus closed, I actually appreciate all the debate uh that that happens on this.
So so I'm I'm very passionate uh about the topic, but I also totally appreciate all of the arguments on the other side.
Um even even you know, though I I probably w lean the other way in some of them, but some of them do inform in in inform some of my views and I and I and I kind of update maybe then to be a little bit more nuanced uh on
my end, but um but you know on the open source UI US side I I generally think that the moneymaker in AI is inference.
Um and so ultimately the dollars are gonna flow to the infrastructure stack one way or another.
I think in a world where you only had one or two labs, um, and somehow there were these in insanely kind of closed secrets about training, um, and and you had like real intellectual property that was protectable and patentable and nobody ever could know about and it's like, you know, totally locked down.
Then in that world, I think you could probably argue that like, you know, there's another couple layers of the stack that you could kind of close off.
But in a world where we're gonna have, you know.
three to five players in the US that consider it sort of existential to them to have leading models.
Then then you have enough of a competitive dynamic in the market where you have to expect that that the cost of tokens converge closer and closer to the to the cost of the infrastructure over time.
Not to zero.
Um I'm I'm definitely not a believer that there's no margin there, but but closer.
So, you know, in the 20, 30, 40% range uh on top of the cost of infrastructure, which is different from 70 or 80 or 90%, let's say.
So if that's
That's the case, then actually if you had an open model and you powered, let's say, the preferred infrastructure for that open model, or you created the post-training environment for that open model, or you were the kind of considered the safer brand for deploying that open model by enterprises, I actually would argue make almost the same amount of revenue just by again driving the inference of that model.
And so if you assume that with things like, you know, I I would actually make the case that even for somebody
Like an open AI, if you fast followed your your frontier models with open source versions of, let's say, the prior generation at a more consistent, you know, pace, I actually think you would you would uh be even more competitive with your frontier models because it would keep more and more of the use cases within your ecosystem and your family.
Um, and you probably could just power a lot of the inference of those open models as well.
Showing 101–120 of 937 · page 6 of 47
← Previous
Next →