Evidence receipt / evaluation
Published · transcript-backedShawn Wang: evaluation
19 Jul 2023 Latent Space Llama 2: The New Open LLM SOTA (ft. Nathan Lambert, Matt Bornstein, Anton Troynikov, Russell Kaplan, Whole Mars Catalog et al.)
“Because open AI will not randomly deprecate your models on you, you know, every three months. And I do think that for people who want a certain level of stability and are okay with trading off not being state of the art in three months I think that is a perfectly reasonable tradeoff.”
Source trail
Everything needed to verify it.
- Speaker
- Shawn Wang
- Attribution
- Verified speaker
- Claim type
- evaluation
- Recorded
- 19 Jul 2023
- Publisher
- Latent Space
Transcript context
…hope for it and that's a huge problem, right? Falcon, if you ask Falcon about human rights abuses in, in the Middle East, it has some very different opinions and I want to understand why. I want to know how they got it to, to do those things. Yeah, yeah, exactly. Yeah, we won't know. And we can, all, we can, all we can do is ask for more transparency there. But I do, I do support the you know, the concepts of building a business on open source models. Because open AI will not randomly deprecate your models on you, you know, every three months. And I do think that for people who want a certain level of stability and are okay with trading off not being state of the art in three months I think that is a perfectly reasonable tradeoff. Okay. I wanted to go back to Nathan A. Little bit and talk a little bit more about the preference data and the R R L H F data. So you estimated a 25 million cost for LAMA two. And as far as I can tell, That's, that's actually primarily data collection, not GPUs. Yeah. This is based on kind of our pilot contract to do preference data collection at hug and paste cuz we can give, like we're collecting a small amount of data in a similar way and if you do a back of the envelope cost calculation and scale it up by whatever, like 10 or a hundred x that what they did, then you get towards this 20 million number and it could be higher depending on how many flags they end up using in their data. So I think what they did was safety is pretty interesting. So they like separated it and collected metadata and that means they could also collect other metadata during the process. And as you kind of add more knobs to the preference data collection because it takes longer for people to do the task and the cost goes up. So I think pretty safe to say order of 10 million, especially given, because that's what was rumored with open AI around ChatGPT and everything like that. So, It is not a shock at all to me. And, and is the…
Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.