Evidence receipt / belief
Published · transcript-backedSoumith Chintala: belief
6 Mar 2024 Latent Space Open Source AI is AI we can Trust — with Soumith Chintala of Meta AI
“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.”
Source trail
Everything needed to verify it.
- Speaker
- Soumith Chintala
- Attribution
- Verified speaker
- Claim type
- belief
- Recorded
- 6 Mar 2024
- Publisher
- Latent Space
Transcript context
…Yeah, one of my friends, Carlo at Berkeley, he worked on a project called M3L, which is two CNNs, one for tactile feedback and one for image. When you say hardware, is it running all these things on the edge or is it just like the actual servos and the- By hardware, I mean the actual servos, like the motors, servos, even the sensors. 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. We have, our skin is all available touch sensing and we have some of the most efficient, some of the most high capacity motors that can lift large loads in the dexterity of a hand and stuff. So in terms of hardware, I mean in terms of those capabilities, we haven't figured out how to do a lot of this stuff. I mean, Tesla has been making incredible progress. One X, I think announced their new thing that looks incredible. Some of the other companies figure and others are doing great work. But we're really not anywhere close to the hardware that we feel like we need. And there's obviously the other thing I want to call out is a lot of what people show works, but has to be fixed all the time. And like, that's the other thing we are incredible at. Like we don't need any maintenance or the maintenance is part of us. If you buy a product, electronics product of any kind, you buy a PS5, you don't say, oh yeah, my PS5 breaks every six days and I have to do some reasonable amount of work on it. But that's robotics. Like if it's not industrial robotics where it's very controlled and specialized or whatever, you're talking about reliability in those ranges. So I think people don't talk about the reliability thing enough. Like what I mean, we're going to enter the commercialization phase. I mean, we're going to start thinking about, okay, now we have this thing and we need to figure out how to get reliability high enough to deploy it into homes and just sell it to people and Best Buy or something. So that's the other factor that we have to make a lot of progress on. I just realized that Google has a play in this with Palm E and stuff and OpenAI obviously has a long history of doing this stuff. Is there anything at Meta? No robotics stuff in Meta?…
Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.