Public evidence record
Published podcast speaker
Claim ledger
30 transcript-backed records
01 / belief
“I think I just happened to like something that is now one of the coolest things in the world or whatever.”
02 / belief
“You know, we're going at it from VR and AR and, you know, we showcase a lot of that stuff. I think for Meta, the robot is not as important as like the physical device.”
03 / belief
“Everything else, like actually finding extrinsic success and all of that, I think is the way I think about it is like somewhat immaterial.”
04 / belief
“I think the more near-term effects are obviously going to be around things that provide more obvious utility in the short term, like maybe smelling cancer or repelling mosquitoes better, or, you know, stuff like that.”
05 / belief
“If you don't have enough compute, you figure out how to make do with smaller models. But no one as of today, I think would feel like they have enough compute.”
06 / belief
“I don't really have a lot of calls to action, because usually I think people should be intrinsically, like, figuring it out.”
07 / belief
“I think we have incredible vision that's still it's so much better compared to in the field of view and in resolution compared to any of the cameras we can buy.”
08 / belief
“Like eventually that's where it will land. And I think there they will kind of fail on the lack of differentiation.”
09 / belief
“I think the moat that NVIDIA has right now, I feel is that they have the interconnect that no one else has, like AMD GPUs are pretty good.”
10 / belief
“Like I think Together and Fireworks and all these people are trying to build some faster CUDA kernels and faster, you know, hardware kernels in general.”
11 / belief
“My view is actually hardware is still the bottleneck and AI is also a little bit of bottleneck, but I don't think there's any obvious breakthroughs we need.”
12 / uncertainty
“Like you wouldn't wanna be doing something so obscure that people are like, I don't know, like you can work on it.”
13 / commitment
“Yeah, I think the performance pitch for Mojo was like, we're going to be performant even if you have a lot of custom stuff, you're going to write arbitrary custom things and we will be performant.”
14 / belief
“There was OPT before, which I think I'm also very proud of because we bridged the gap in understanding of how complex it is to train these models to the world.”
15 / belief
“We probably, I think, didn't train it for long enough, right? That's kind of obvious in retrospect.”
16 / observation
“You understand, the key is you create variations of your documents and you know how to do that because you have a symbolic model or like some implicit symbolic model of language.”
17 / evaluation
“Our own silicon, I think we've even showed the standard photograph of you holding the chip that doesn't work.”
18 / belief
“Instead of answering specifically about that Llama 2 situation or whatever, I'll tell you how we think about things.”
19 / belief
“I think you're basically trying to, what you're doing is you're saying, well, language, I know how to parametrize language to an extent.”
20 / belief
“I mean, I think the details would add more spice, but also wouldn't add more understanding.”
21 / belief
“I think at that point you would be investing in them for their expertise of one kind.”
22 / belief
“I think there's this entire subfield where they're trying to figure out these answers called like the science of deep learning or something.”
23 / belief
“I think all the near-term stuff, I think if he's able to figure out a near-term value for it, they, as a company, can sustain themselves to then eventually try to make progress on the long term, which is really in uncharted territory.”
24 / preference
“We very much don't want to king make the hardware side of things. So as the MacBooks have GPUs and as that stuff started getting increasingly interesting, we pushed Apple to push some engineers and work on the NPS support and we spend significant time from Meta funded engineers on that as well because a lot of people are using the Apple GPUs and there's demand.”
25 / evaluation
“I used to do image models and stuff and I actually stopped doing GANs because GANs were getting so hot that I didn't have any calibration of whether my work would be useful or not because, oh yeah, someone else did the same thing you did.”
26 / evaluation
“That was my main criticism. And I'm pretty sure if before they released it, they showed it to their other stakeholders who would be caring about this benchmark because they are present in it, they would have easily just pointed out these gaps.”
27 / evaluation
“The only way we have to teach them is generating a bunch of inputs and outputs and gradient dissenting over them. So in areas where we have the symbolic models and we need to teach all the knowledge we have that is better encoded in the symbolic models, what we're doing is we're generating a bunch of synthetic data, a bunch of input output pairs, and then giving that to the neural network and asking it to learn the same thing that we already have a better low rank model of in gradient descent in a much more over-parameterized way.”
28 / preference
“Like if a better open source model comes, really no one says, ha, I don't want to use it because there are ecosystem effect, I'm logged into my ecosystem or, I don't know if I like the models, you know, whatever.”
29 / recommendation
“Very interesting, but separate times. But to answer your question, Alessio, I think mainly Lepton, Fireworks are the two most obvious ones, but I'm sure the fingerprints are a lot wider.”
30 / prediction
“At some point MLX either says we will only be supporting Apple and we will just focus on enabling, you know, there's a framework if you use your MacBook, but once you like go server side or whatever, that's not my problem and I don't care.”