Dylan Patel

The next thing to do for open source post-training is to scale up verifiers, to scale up data, to replicate some of DeepSeq's results. And it's awesome that we have a paper to draw on and it makes it a lot easier. And that's the type of things that is going on among academic and closed frontier research in AI.

This goes very back to the license discussion. So DeepSeq R1 with a friendly license is a major reset. So it's like the first time that we've had a really clear frontier model that is open weights and with a commercially friendly license with no restrictions on downstream use cases, synthetic data, distillation, whatever.

This goes very back to the license discussion. So DeepSeq R1 with a friendly license is a major reset. So it's like the first time that we've had a really clear frontier model that is open weights and with a commercially friendly license with no restrictions on downstream use cases, synthetic data, distillation, whatever.

This goes very back to the license discussion. So DeepSeq R1 with a friendly license is a major reset. So it's like the first time that we've had a really clear frontier model that is open weights and with a commercially friendly license with no restrictions on downstream use cases, synthetic data, distillation, whatever.

This has never been the case at all in the history of AI in the last few years since ChatGPT. There have been models that are off the frontier or models with weird licenses that you can't really use them.

This has never been the case at all in the history of AI in the last few years since ChatGPT. There have been models that are off the frontier or models with weird licenses that you can't really use them.

This has never been the case at all in the history of AI in the last few years since ChatGPT. There have been models that are off the frontier or models with weird licenses that you can't really use them.

So this goes to what open source AI is, which is there's also use case restrictions in the Lama license, which says you can't use it for specific things. So if you come from an open source software background, you would say that that is not an open source license. What kind of things are those, though? At this point, I can't pull them off the top of my head.

So this goes to what open source AI is, which is there's also use case restrictions in the Lama license, which says you can't use it for specific things. So if you come from an open source software background, you would say that that is not an open source license. What kind of things are those, though? At this point, I can't pull them off the top of my head.

So this goes to what open source AI is, which is there's also use case restrictions in the Lama license, which says you can't use it for specific things. So if you come from an open source software background, you would say that that is not an open source license. What kind of things are those, though? At this point, I can't pull them off the top of my head.

It used to be military use was one, and they removed that for scale. It'll be like... like CSAM, like child abuse material. That's the type of thing that is forbidden there, but that's enough from an open source background to say it's not an open source license. And also the Lama license has this horrible thing where you have to name your model Lama if you touch it to the Lama model.

It used to be military use was one, and they removed that for scale. It'll be like... like CSAM, like child abuse material. That's the type of thing that is forbidden there, but that's enough from an open source background to say it's not an open source license. And also the Lama license has this horrible thing where you have to name your model Lama if you touch it to the Lama model.

It used to be military use was one, and they removed that for scale. It'll be like... like CSAM, like child abuse material. That's the type of thing that is forbidden there, but that's enough from an open source background to say it's not an open source license. And also the Lama license has this horrible thing where you have to name your model Lama if you touch it to the Lama model.

So it's like the branding thing. So if a company uses Lama, technically the license says that they should say built with Lama at the bottom of their application. And from a marketing perspective, that just hurts. I could suck it up as a researcher. I'm like, oh, it's fine. It says Lama Dash on all of our materials for this release.

So it's like the branding thing. So if a company uses Lama, technically the license says that they should say built with Lama at the bottom of their application. And from a marketing perspective, that just hurts. I could suck it up as a researcher. I'm like, oh, it's fine. It says Lama Dash on all of our materials for this release.

So it's like the branding thing. So if a company uses Lama, technically the license says that they should say built with Lama at the bottom of their application. And from a marketing perspective, that just hurts. I could suck it up as a researcher. I'm like, oh, it's fine. It says Lama Dash on all of our materials for this release.

But this is why we need truly open models, which is, we don't know DeepSeek R1's data.

But this is why we need truly open models, which is, we don't know DeepSeek R1's data.

But this is why we need truly open models, which is, we don't know DeepSeek R1's data.

hell yeah that's that's what i'm saying and yeah and that's why it's like we want this whole open language models thing the olmo thing is to try to keep the model where everything is open with the data as close to the frontier as possible so we're compute constrained we're personnel constrained we're