What's the monetization model here? Is this a closed-source version of their open-source model? (That's suggested by the phrase in the article, "a commercial version of LLaMA, its open-source large language model".)
Like others said it’s probably to commoditize their competition. The models don’t matter so much as ownership of the platform and critical data. Which is why OpenAI is in a tricky position (although I guess they’re partnered with Microsoft).
It seems like the existing large platforms of today—Microsoft’s enterprise moat, Google’s ads and internet services, Meta’s social networks, Apple’s consumer and mobile products—will remain the primary platforms of the future. So having models that can operate exclusively on those platforms via integration to their key products and date will only continue this trend. If you’re an outsider with an AI model, you’ll have a harder time getting access to critical data and your standalone AI product (e.g., ChatGPT) won’t be as useful.
More broadly speaking, I believe the days where the top X largest companies in the stock company would be displaced by newer companies every decade or so is over. The FAANGs just control so many major platforms in so many aspects of our lives.
> More broadly speaking, I believe the days where the top X largest companies in the stock company would be displaced by newer companies every decade or so is over. The FAANGs just control so many major platforms in so many aspects of our lives.
It also helps that they buy or otherwise cooperate to destroy their competition in questionable ways while heavily lobbying the gov to favor them over others in a quid-pro-quo that benefits politicians and not their constituents.
> More broadly speaking, I believe the days where the top X largest companies in the stock company would be displaced by newer companies every decade or so is over.
I disagree: I think big tech is hard to disrupt ATM because the companies are still young and nimble. In the last cycle, the companies being displaced were ancient (by tech standards). When Google and Facebook are 30 years old, their DNA will get in the way of adopting to a new paradigm that will change the world. A paradigm that may be to the Metaverse what the smartphone was to the Apple Newton
Google opensourced Tensorflow because they believed it would help with the hiring process: if researchers could use the same framework to do their PhDs as Google used in their production systems, that was seen as an advantage.
Maybe that's Meta's play here? Maybe the idea is that the ecosystem around a model could be as valuable or more valuable than the model itself too, so an OSS model could benefit Meta a lot more by gaining more of the ecosystem mind share?
Or Maybe Yann LeCun is just a hippie that dreams of free love, hard drugs and open-source models?
LLaMA is already in it's way to becoming an industry standard (in my opinion, look at llama.cpp plus everything build on LLaMA). There are benefits to being able to set direction like that. Same as pytorch for example, it's not just about direct revenue it's about everyone building on and contributing to your platform.
They might have done well to make gg an offer he couldn't refuse and take on ggml and llama.cpp as an open source project.
LLaMA isn’t licensed for commercial use. It’s probably an update to the licensing.
Facebook benefits heavily from the open source development done on LLaMA. There was a report I saw that facebook has started using llama.cpp internally for inference. Updates to the licensing will cement facebook as the go to choice for open source language models.
Based on the podcast with Lex Friedman and Mark Zuckerberg, see ~minute 30.
My hypothesis based on the context of Mark discussing the release is that it's going to be completely open source and can licensed to be used commercially. Not that Meta is going to add a whole new revenue side of business to compete with OpenAI. i.e. "Here is model, with commercially permissive licensing" not "Here is model that you can use commercially but must pay me"
Meta has been one of the major open source contributors for about a decade now. They open source/contribute to a lot of tech, as their business isn’t about tech, but products.
This isn't some recent revelation or anything. Facebook's AI team (FAIR) open sourced their major technology in 2017 with Pytorch. In 2018 they published Pytext in an age when most people didn't know what a Large Language Model even meant. Seeing LLaMA get made should not be a surprise to anyone who is familiar with the history of AI research. It's like hearing people call CUDA an "unfair advantage" while ignoring billions of Nvidia R&D dollars getting spent in the AI sector over the course of a decade.
It might feel like "brand rehab" or "good will" as a consumer, but a lot of this work was put in motion a while ago.
I don’t know, maybe they don’t need to monetize their model? I don’t know if they have to, they need their models to be the best and to support their core business of ads, anything that keeps users on their platform for any reason is their goal. They need their models to be an industry standard and one upon which other things are built.
That's a huge reason to do it also, but it also makes sense if you have researchers + developers improving the engine of something that powers your product. The moat / competitive advantage at FB is their network, not so much the proprietary underlying tech.
People often say this but having interviewed ~200 facebook engineers over the years, their scaling tech around both software and hardware is pretty impressive.
Yeah I guess it's a competitive advantage when a competitor (Twitter) is showing to have technical problems operating at global scale with a smaller team. Their scale is not trivial by any means. But people aren't going to go to FB because they have the best LLM, makes sense to offload that development to the open source community.
You still need to build real-time serving infrastructure on top of LLaMA/Vicuna/Alpaca in order to compete with ChatGPT/OpenAI so it's not going to be done by that many companies and OpenAI already has a mindshare/first mover advantage.
When you use ChatGPT you are leasing their GPU infrastructure and their proprietary model, this opens the possibility of leasing GPU infrastructure from another company and using an open source model. You don't necessarily need to do the hard parts yourself, you can hire it out to competing companies.
Sure, but it's extra work slowing you down as your competitor is surfing the wave at full speed. Moreover, you are relying on an old LLM whereas OpenAI is developing newer versions of theirs, keeping their competitive advantage. Even Google who has the infra has a ridiculously bad LLM to compete.
Well, if they really released it as open source, I guess depending on the exact license a company that modifies(fine tunes) it and wants to make money on that modified version would have to distribute the weights and/or disclose the details about how they fine tuned it. On what data etc. By offering a commercial license , the buyer can do anything they want.
Meta is a company that makes money off of users endlessly browsing content. It would follow that making it easier/faster to generate content would benefit Meta.
