Google and HuggingFace Partner in Boost to Open Source AI

0:01Today on the AI Breakdown, we're looking at a new partnership between Hugging face and Google. Before that on the brief, the FTC has officially opened up antitrust investigations around the deals between Anthropic, Google, Amazon, and Microsoft and OpenAI. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information about our YouTube, our newsletter, and our Discord. Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five minutes. Recently, we had shared reporting that suggested that there was a lot of interest among big government agencies investigating whether some of these partnerships in the artificial intelligence space were in violation of antitrust laws.

0:45In fact, it appeared that there was a bit of a territorial match, but that finally seems to have been resolved. The FTC is officially launching a probe of how investments between companies like Microsoft and OpenAI and Amazon and Anthropic impact competition. Now, it's important to note that so far, there is no accusation of wrongdoing. This isn't the FTC saying that the deal between Microsoft and OpenAI was an antitrust violation, or the deals between Google and Amazon and Anthropic are antitrust violations. But at the same time, the organization has expressed concern that big tech has an unfair advantage when it comes to this increasingly important field.

1:22According to the Wall Street Journal, the FTC said it would use the study to probe the company's investments and partnerships and how they affect the competitive landscape for AI. FTC Chair Lina Khan said, Our study will shed light on whether investments in partnership pursued by dominant companies risk distorting innovation and undermining fair competition. The piece from the Wall Street Journal also affirms that the FTC had been in fact negotiating with the Justice Department around how to divide this work, although it seems that progress is now being made. Now apparently the FTC has been a little bit nervous about this ever since the big Microsoft investment in OpenAI.

1:54They were apparently frustrated and concerned that they hadn't been told about that investment in advance. Now, part of what makes this complicated is that these partnerships reflect not just big tech being aggressive, but big tech being the only institutions with deep enough pockets to actually fund these projects on the scale that they cost. There aren't venture funds, for example, out there that could match Microsoft's$13 billion investment into OpenAI. They just don't exist. The economics of venture funds don't work at that scale. At the same time, OpenAI clearly isn't ready to go public, and so there's a very limited set of options.

2:26Now, when it comes to the argument against this sort of inquiry, Adam Kovacevic, the chief executive of the Chamber of Progress, said that the FTC's focus on Amazon, Google, and Microsoft, as well as OpenAI and Anthropic, shows that the FTC has, quote, embraced the idea that the AI space is already locked up and views these investment deals as evidence of that. Kovacevic thinks that that's patently absurd, saying every single company is talking about what they will do with AI. Now, in terms of the companies themselves, Google said in a statement that it's not locking the companies that it invests in to work with competitors.

2:56Case in point, both Google and Amazon have invested billions of dollars into Anthropic. Microsoft also commented saying that their partnership, in fact, promotes competition and accelerates innovation. But whatever the case, this investigation is now happening, and it shows just how important AI has gotten in Washington. Speaking of which, Politico today published a piece called The Campaign to Take Down the Biden AI Executive Order. Republicans, right-of-center think tanks, and tech lobbyists are working to defang the industry-facing sections of Biden's AI executive order. Now, in some ways, this isn't just a story about AI.

3:29It's a story about the way that executive power is wielded in the U.S. right now, which is in of itself a story about the power realignment between the executive and the Congress in a world where Congress can never get anything done. But the key note is that the Biden administration has argued that AI represents a national security threat, which warrants using the Defense Production Act, which gives the federal government powers over parts of the private sector. This group of Republicans, however, say that's overreach. Now, there are a few ways that folks are trying to combat this. There are some number of lawsuits from private sector think tanks, but Politico also reports, quote, Republican Senate commerce staff are also reportedly slowing down all AI regulation passing through their committee.

4:06Senate Republican Whip John Thune told reporters he was working with Ted Cruz and other Republican lawmakers to curb the DPA's ability to create new AI testing and training protocols for the tech industry. AI may not be an election issue yet, but it is definitely political. Moving back over to the private sector for a moment, the Financial Times is reporting that Elon Musk and his ex.ai are seeking to raise$6 billion at a$20 billion valuation. Says the FT, Elon has courted wealthy individuals and investors around the world in recent weeks according to multiple people familiar with the matter. These talks have included family offices in Hong Kong.

4:38Another person said that Elon had also targeted sovereign wealth funds in the Middle East, but some said that at this stage, Elon is testing to see how much investor appetite there is for numbers at this scale. Going back to the reality of the costs of operating in this market that we were discussing earlier, the FT writes, the scale of the attempted fundraising reflects the enormous costs required to develop generative AI, which requires huge computing power, vast amounts of data, and cutting-edge chips. So whether Elon ends up fundraising right now or not, it seems very clear that he is set on having a war chest to actually have X.AI and Grok compete.

5:10Anyways, friends, that is going to do it for today's AI Breakdown Brief. Next up, the main AI breakdown. Before we get to the main part of the episode, one quick note about the AI education beta that you have been hearing about all week. This is the learning program we've been running for a couple months now that's all about tutorials, challenges, basically getting you learning AI by actually using AI faster, more effectively, more efficiently, and hopefully in a way that's more fun too. If you sign up, you'll get access to a library of 60-plus tutorials, case studies, and challenges, new lessons that drop every weekday, and a passionate community of like-minded learners.

5:46We cover everything from LLMs and prompt engineering to image generators, voice synthesizers, AI no-cold tools, as well as professional applications like presentation generation, website building, and more. Due to a bunch of people saying that the window for enrollment was too short, I'm actually going to extend it to Sunday night at 1159 East Coast time, So you have a couple more days to register. Go to bit.ly slash AI beta to learn more and sign up. I'd love to see you there. But now let's get back to the show. Welcome back to the AI Breakdown. Today we got some news that was very exciting to the open source advocates within the AI community.

6:25Google and Hugging Face announced a major new partnership, specifically between Hugging Face and Google Cloud. Philip Schmidt from Hugging Face tweets, we will collaborate with Google to foster open AI innovation across open science, open source, cloud, and hardware. Why this matters? Keeping AI open, accessible, and efficient. So before we get too deep into this, let's not assume that everyone knows exactly what Hugging Face is. And by way of summing up, let's turn to our favorite research tool, Perplexity. I asked Perplexity, what is Hugging Face? And it describes it as a French-American company that provides a machine learning and data science platform and community.

7:00It is designed to help users build, train, and deploy machine learning models, offering infrastructure for running AI applications. The platform is known for its open-source transformers library, which is widely used for NLP tasks. Hugging Face is often referred to as the GitHub of machine learning because it allows developers to share and collaborate on machine learning models and datasets. So this is a company that's been around since 2016 and has become a hub for, as it said, AI researchers and developers to share what they are working on, to find open-source models, to run them. often you will see, for example, when I'm talking about some new tool, that the demo that I'm running and sharing with you guys is hosted on Hugging Face.

7:37Now, one of the other important things to know about Hugging Face is that it has positioned itself as one of the pillars of the open source AI movement. Once again, the way the perplexity sums this up is that Hugging Face's commitment to open source collaboration has catalyzed innovation in natural language processing, allowing for communal growth and development of the technology. Hugging Face has positioned itself as a hub where data scientists, researchers, and ML engineers converge to exchange ideas, seek support, and contribute to open source initiatives. Now, a couple of examples include their Transformers library, as well as them releasing open source serving solutions for LLMs and an open source chat UI for LLMs that are tools that tons and tons of companies out there use.

8:15So, key takeaway here is that Hugging Face has a long-term demonstrated commitment to the open source side of AI. So, let's talk about what this actual partnership includes. On the Google Cloud blog, they write, Google Cloud and Hugging Face announced strategic partnership to accelerate generative AI and ML development. Developers will be able to train, tune, and serve open models quickly and cost-effectively on Google Cloud. So basically, the partnership at core is one that will, quote, allow developers to utilize Google Cloud's infrastructure for all Hugging Face services. It will enable training and serving of Hugging Face models on Google Cloud.

8:48Now, it is explicitly about a commitment to open AI development. The partnership, they write, advances Hugging Face's mission to democratize AI and furthers Google Cloud support for open source AI ecosystem development. With this partnership, Google Cloud becomes a strategic cloud partner for Hugging Face and a preferred destination for Hugging Face training and inference workloads. So basically, in some ways, Google Cloud is trying to become an infrastructure layer for open source AI via this partnership with Hugging Face. Again, they write developers will be able to easily utilize Google Cloud's AI optimized infrastructure, including compute, tensor processing units, and GPUs to train and serve open models and build new generative AI applications.

9:27So some of the specifics include connecting Hugging Face models with Vertex AI, which they describe as Google Cloud's purpose-built end-to-end ML Ops service. They want to give more open source developers access to Cloud TPU version 5e. But overall, it feels to me like Google is really trying to establish its place in this overall open source space. Now, you may remember last year, one of the big interesting and revelatory documents was a memo leaked from inside Google by someone who worked on AI inside that company who argued that OpenAI wasn't Google's true competition in the AI space, but instead it was the open source movement.

10:02They talked about how, ironically, because of the leak around Meta's Llama 1 model's weights, Meta, Mark Zuckerberg's company of all things, had found itself at the very center of a radical shift in open source AI development. The Googler at the time noted that open source developers were making surprisingly quick progress on problems that Google had identified as significant for the space and thought it was going to take a lot longer to solve. And why this is relevant, of course, is that as all of these big tech companies try to position themselves in the larger AI game, their relationship with open source, or contra it in the case of, for example, OpenAI and Microsoft, is one of the key pieces of positioning.

10:40Meta has played this card very well and has become a central piece of that ecosystem. Although even Meta has found itself recently competing more aggressively because of the rise of Mistral, which has created sort of a renegated upstart competition with models that developers really love. So coming back to the Google Cloud and Hugging Face partnership, again, part of it reads as a positioning play for them. Now, Jeff Boutier from Hugging Face, who does product and growth, joined Brad Smith at Yahoo Finance for an interview and talked a lot more about what the partnership was and what it wasn't.

11:11Now, the first question that Smith asked was, is this Google Hugging Face deal a can or analogous to the Microsoft OpenAI deal. Jeff says, no, I would say it's quite different. And it's really an opportunity to make it easy for Google Cloud customers to build their own AI with open models. Today, we host over a million models, datasets, applications for text, audio, video, time series, biology, you name it. And all of these models will be very easy to access from Google Cloud. Will be very easy to use with the latest hardware available on Google Cloud. So already we have an interesting contrast.

11:41Whereas Google Cloud's announcement about this was all about how they were supporting open source developers. in other words, developers that were already in that space, Jeff and Huggingface are basically saying that this alliance actually allows more existing Google Cloud customers to become part of that open developer community. Now, in addition to details of the partnership, Smith also asked about some of the underlying principles. He said, One of the interesting comments that we heard from the CEO of Salesforce, Mark Benioff, during our conversation with him at Davos at the World Economic Forum, was talking about how AI should be a human right.

12:11He also did address some of the concerns. What concerns still linger from your perspective around generative AI that the industry still needs to solve for. Jeff responds, The open source community has been tackling all of the main issues that have been identified by the AI research community and working through those issues openly, transparently through open source. And I think that's a huge part of the solution. And what we want at Hugging Face is also to make those open models and open source AI easily accessible to companies so that they control their own AI destiny so that they can host themselves, the models, and they can protect their data as well.

12:40Now, of course, this gets into a much larger conversation around open source and its role with AI. On the one hand, AI safety advocates are hugely concerned about open access to advanced AI models. The argument tends to follow some version of, do we really want terrorists, for example, to have access to technology that can fabricate new types of weapons, especially biological or chemical weapons? And indeed, this isn't just some fringe view. Obviously, the folks at OpenAI tend to have it as well, and have said explicitly that they believe that their previous take on open source was incorrect. Another way to put that is that there is a set of people who believe that the risks of making super advanced AI freely available in an open source way are greater than the risks that come with concentration of advanced AI in the hands of a small number of companies.

13:26There is of course then the opposite view, that the biggest risk is not rogue elements getting access to advanced AI, but advanced AI being controlled by just a small number of people. Ironically, given his 20-year history, Zuckerberg has become part of the headliners of that second perspective. Most recently, he made news when he announced that Meta's new goal was not just to create advanced AI and release it open source, but to actually create artificial general intelligence that could be released open source. Now, he didn't go so far as to commit 100 % that if they achieved AGI, they would release it open source, but only because he said, who knows what we'll discover, and we need to leave ourselves some room.

14:03I don't want to be trapped by things I say now, but it's very clear that this is the direction they're moving. Jan LeCun, the chief AI scientist at Meta, also is one of the loudest and most vocal advocates for this. After Zuckerberg's announcement, Aravon Srinivas from Perplexity wrote, open source AGI is an amazing vision. You are building a very powerful technology and actually aligning it to what makes sense for the world. More people have a say in what makes sense and doesn't. Zuck and Jan LeCun are leading the revolution, and it's going to have an incredible short-term impact for the AI ecosystem as a whole.

14:31Meta gets a lot of crap, but there's literally no other company doing this today. But then again, this pronouncement brought articles like this one from The Guardian that reads, Very scary. Mark Zuckerberg's pledge to build advanced AI alarms experts. Meta CEO accused of being irresponsible by considering making tools on par with human intelligence open source. Now, one of the issues that people have, basically with all of these debates, are how many of these decisions are being made in the boardrooms of extremely powerful companies rather than out in the open. Dr. Andrew Rogoisky, for example, a director at the Institute for People-Centered AI at the University of Surrey said, there are deep and complex arguments about the merits of open source and current AI models.

15:07Pushing that into the realm of AGI could be world-saving or catastrophic. These decisions need to be taken by international consensus, not in the boardroom of a tech giant. Now, one thing I will note along those lines is that I watch this very closely. And people like Sam Altman, who are much more on the open source could be very bad for us side of things, have also said things like this, that there needs to be a broader set of stakeholders involved in these types of decisions. However, so far that group hasn't really advanced what that could meaningfully look like. And in the absence of some vision to rally around, it defaults back to the companies that are building it.

15:40Now, if you're interested in staying at the edge of this open source AI debate, let me give you two Twitter account recommendations. First, Jan LeCun, who I just mentioned, at Y-L-E-C-U-N on Twitter slash X. He is always commenting on these issues, always resharing every post that relates to them. And so you will definitely get the pro open source side from him. On the flip side is the at AI safety memes account, which is literally at AI safety memes. The account's bio reads techno optimist, but AGI is not like other technologies. This account tends to do a lot of curation of these discussions.

16:12and while it comes from a very discreet perspective, it does so with extremely good faith. So if you're looking to really understand that side and that set of concerns, you can't do much better than starting there. You'll also see a lot of other accounts that you might be able to follow. As always, my recommendation is to listen to everyone and think deeply about these issues because they're going to impact us all. For now, however, that is the story for today. Hugging Face and Google have teamed up. Google is planting a bigger flag in the open source AI space and this whole world continues to be interesting.

16:42Thanks for listening or watching as always, and until next time, peace.