Searching...
17 results for “ai rollups”
“Why AI-rollups must force standardization to succeed where traditional buy-and-build fails”
fit your product? That's amazing question. That's exactly the latter. And I think that's exactly where the difference between your typical kind of buy and build and what we're doing is coming up. Bec
22:29 / 23:04
contrarian take22:29 – 23:04
So Charles Rolllet, that's r o l l e t one. That's my handle on x. Ah, okay. So so there's another Charles Rolllet out there. Yeah. There's an imposter. Actually, Charles Rolllet is not an uncommon name in French. There's a bunch of us. So make sure
20:59 / 21:36
from transcript20:59 – 21:36
“How data reveals the optimal operational processes by ditching industry best practices”
the selection of the best practices of operational processes across The UK with not just selecting the best one of the variety, but also quite often selecting none of the existing and reviewing all of
23:40 / 24:15
framework23:40 – 24:15
Rippling handles all the can't get it wrong admin work of payroll and benefits, giving you back hours every week. But it does a lot more than that. Rippling is a game changer for the entire company, with tools for HR, IT, and spend, all built from th
3:44 / 4:46
from transcript3:44 – 4:46
But I'm excited and glad that you were able to hop on the mic and kinda go over this with me. No. I mean, that it's great talking to you, Marianne. And, yeah, you've done great coverage on this case as well. And it can be really hard to keep track of
20:14 / 21:16
from transcript20:14 – 21:16
...rollups and automate those. Getting them to understand what agents are sometimes can be, you know, an obstacle course, but those things are happening. Right? People see them as opportunities. Coming up, Mark answers some of your questions, and we div
17:16 / 18:02
from transcript17:16 – 18:02
And we're papering over the limitations, and we're kind of working around them in all sorts of ways, whether it's the and the LLM itself and the training data or and the infrastructure around and everything that we're doing to make them work. But tha
6:46 / 7:48
from transcript6:46 – 7:48
Dario and Sam had prior to this moment, or at least the way I heard them, made it sound like they were just gonna build the next biggest thing and get the same amount of gain. They had left that impression. And so we get to this place as you describe
32:15 / 33:17
from transcript32:15 – 33:17
less famously for also showing that you can scale reinforcement learning training and get kind of this log x axis and then a linear increase in performance on y axis. So there's kind of these three axes now where the traditional scaling laws are talk
1:03:41 / 1:04:51
from transcript1:03:41 – 1:04:51
How do you think about what you just said in conjunction with the idea that scaling laws were were hitting this asymptote point? How do you reconcile the two? I don't think there's a lot of debate among senior ML thinkers that we have tremendous room
23:58 / 25:03
from transcript23:58 – 25:03
and serving has these considerations you need to scale. Like, we talked about pretraining. We talked about RL. And then inference time scaling is, like, how do you serve a model that's thinking for an hour to a 100,000,000 users? I'm like, I don't re
1:14:58 / 1:16:04
from transcript1:14:58 – 1:16:04
Get us any amount of compute, we will use it all up and say we still don't have enough. So this is a constraint for the last twenty years or so. But what I'm seeing is with the rise of Gen AI, there are very valuable workloads. For example, AI assist
5:00 / 6:19
from transcript5:00 – 6:19
Representing scaling laws where it became more and more formalized that bigger is better across multiple dimensions of what bigger means. So, and but these are all sort of neural networks we're talking about, and we're talking about different archite
38:57 / 40:02
from transcript38:57 – 40:02
and it it was extremely well suited to scaling it up across many GPUs. At the time we built it, like, many GPUs was tens of GPUs. Right? Like, that was like And what's many GPUs today? Oh, tens of thousands, you know, maybe hundreds of thousands. In
27:09 / 28:21
from transcript27:09 – 28:21
In a GPU, most of the time it's doing inference, it's five or 7% utilized. That means it's 95 or 93% wasted. Over time, I think, as an industry, we get better at things. We can drive the cost of compute down. We can build more efficient data centers
23:12 / 24:18
from transcript23:12 – 24:18
But I I I think we're starting to run up against the limits of of really good data that we can. What's then the problem? So so ultimately, that might mean that, hey. We're training larger and larger models. XAI, again, just just created the largest G
39:41 / 40:53
from transcript39:41 – 40:53
So this can mean just having more text inputs for for your models, but it can also mean things like taking a lot of visual token inputs, image inputs to your models, or generating lots of outputs. And one thing that's been really exciting over the la
3:41 / 4:49
from transcript3:41 – 4:49
Have a podcast?
Run your podcast through Clypt
Get ranked clips, hooks, and ready-to-post copy from your own episodes. Free to try.