Unpacking the Financial Incentives for Open Source vs Closed Source [Markets]

Why Meta is fighting for Open Source LLMs while Microsoft wants to regulate them.

Dec 10, 2023

Hey, it’s Devansh 👋👋

In Markets, we analyze the business side of things, digging into the operations of various groups to contextualize developments. Expect analysis of business strategy, how developments impact the space, and predictions for the future.

If you’d like to support my writing, consider becoming a premium subscriber to my sister publication Tech Made Simple to support my crippling chocolate milk addiction. Use the button below for a lifetime 50% discount (5 USD/month, or 50 USD/year).

Help me buy chocolate milk

With all the attention given to Google’s release of Gemini, you might have missed the updates regarding the EU's AI act. It’s not something you want to miss, b/c the Open Source movement caught a pretty big W-

The legislation ultimately included restrictions for foundation models but gave broad exemptions to “open-source models,” which are developed using code that’s freely available for developers to alter for their own products and tools. The move could benefit open-source AI companies in Europe
—Catch Yann LeCun’s Tweet about this win for Open Source here

This comes around the same time as Meta starting a massive AI Alliance for “Leading Technology Developers, Researchers, and Adopters Collaborating Together to Advance Open, Safe, Responsible AI”. This alliance includes heavy hitters such as AMD, Dell, HuggingFace, IBM, Intel, Oracle, Fast.ai, Stability AI, and NASA.

Notable absentees include Apple, Amazon, Google, Microsoft, Anthropic, and OpenAI, the last 4 of who had started their own Frontier Model Forum a few months ago-

This is very strange behavior. Why do we have two different organizations that want to ensure “safe and responsible AI”, why do they have 0 overlapping members, and why are two groups spending millions lobbying against each other? It wouldn’t shock you, that this is not the ideological disagreement that the companies are pretending this is. In reality, each player’s stance on AI Regulations is tied to their commercial interests. In this article we will be going over the following-

Why Google fell to the enemy within (why they have to go against Open Source).
Why Microsft and OpenAI have been so gung-ho for regulations.
Why Amazon and Apple have been relatively quiet.
Why Meta is lobbying for Open Source (Meta’s stance also explains other Big players who push for Open Source).

Before we proceed, a few readers of mine recommended that I get onto Expert Consulting sites like GLG. US Visa Laws prevented me from doing so, but that has been resolved. I now have a GLG account (and will get others soon). If you’re a client of these sites and would like to book a consultation session, please reach out.

Let’s start with the company that has gotten the rawest end of the situation. The company that not too long ago, looked unstoppable in AI. The Big G itself, Google.

Google AI should fire its entire PR team

The headline might have a few of you raising your eyebrows. All I ask is that you read this with an open mind. Understanding how Google AI’s PR Team has failed them, will help us understand why Google has no choice but to go against Open Source right now.

To summarize what will be a fairly long section- Google AI has been on God mode when it comes to AI Advancements. Technically, they are ahead of the rest of the competition (1.5–2 years in my assessment). However, they have been extremely incompetent at sharing their developments. This significantly undermines their public perception, thereby adding a lot of volatility to their stock. The slightest mistakes cause massive overreactions and massive stock slides. Google’s positioning against Open Source Foundation Models is an attempt to limit volatility by restricting competition to more known entities.

If you’re busy you can skip the rest of this section. But I would recommend reading it to get additional context (and if you read the whole thing, I’ll like you more).

Google’s AI teams have consistently released some of the most groundbreaking work I’ve read. I don’t say this lightly. At this stage, it’s well known that Google invented Transformers, changed the NLP world with BERT, etc. What is not as well understood is just how much more they contributed to the AI Space and how far ahead they were in LLMs.

Google AI shared their publication about the Pathways architecture (their next-gen system for training AI Models) in 2021. That was more than 2 years ago. They were talking about multi-modal AI Models and multi-task training at that stage. Their work with Chinchilla was crucial to scaling Transformers to the current size. They were talking about AI Hallucinations months before most people had heard of OpenAI. If someone were to take just their Flamingo model and build a startup around it today, it would be enough to raise billions of dollars. Their startup would be considered a serious threat to OpenAI and GPT-4. Flamingo was just one of Google’s contributions in April 2022.

Before I committed to writing, I used to make monthly YouTube videos about the 4–5 most important developments in AI. Google AI was invariably mentioned at least once or twice in those updates. Google Researchers and Developers have been on demon-time. By my analysis of Google’s publications, Google is at least 1.5–2 years ahead of everyone in the LLM space.

I picked a random number and went to the Google blog of that number. And this is what I found. I’m telling y’all, that Google Research gets very funky (this blog taught me about something called flood-filling networks, which seems interesting).

This extends beyond just LLMs. Google has released gems in Protein Folding, Robotics Control, Scalable AI, Privacy-Preserving AI, AI for Climate, and Quantum Computing. You could just read the exceptional Google Research Blog and still know about the most cutting-edge developments in the fields. The lunatics that publish there are built different. The fact that OpenAI is mentioned as a serious threat to Google, is criminal and shows you how bad Google has been with highlighting its accomplishments.

All of that is to say that no one, and I repeat no one, deserves to be in the same conversation with Google AI. The only competitor that can claim to have a similar degree of impact in AI is Meta, and they are only in the conversation b/c of their open-source contributions (FAISS, PyTorch…) and for their part in making self-supervised learning more mainstream.

Unfortunately, there have been no efforts to communicate Google’s developments to a wider audience. Google’s primary publicity strategy has revolved around people like me, who interact with primary sources and break them down. This severely restricts the audience because there are very few people who can engage with these sources and even fewer who will (many of you have the technical knowledge to read the papers, but are too busy to do so). When it comes to influencing public opinion, this is a really bad strategy.

This brings a double whammy of trouble for Google- the achievements of their competitors are lionized, while any mistake they make is demonized. Their release of Bard should have a real “daddy’s home” moment. Instead, all the press was about Bard’s costly mistake.

I was one of the few who held that Bard would beat GPT after this debacle. Most people didn’t understand how far ahead Google was.

We see something similar with Gemini. The top stories are not pretty-

Compare this to the ballistic reception that every little OpenAI update gets. The dev day had most people losing their minds, even though there wasn’t much new that was shared. Nvidia and Sam Altman’s Chip Investments both generated a lot of positive public attention. Google’s long-time AI investments in TPUs are not nearly as well known, even though TPUs are objectively better for AI.

In my post about how Social Media enforces conformity, I talked about how a deluge of repetitive content can reinforce false narratives. Google has found itself at the negative end of this. Their terrible media strategy has reinforced the narrative that they are falling behind in LLMs, and now everything they do is painted as them catching up. This makes any mistakes they make much more unforgivable than the ‘innovative’ OpenAI.

Google is in a really good place. They just can’t show that to the public, so their stock takes a massive beating every time. They don’t have any threats to their core business (even if you make better AI than Google, building products to fight them is a nightmare). But as long as people believe they do, they will have to keep dancing to other people’s tunes. Google learned this the hard way when AI Experts started proclaiming the death of search after ChatGPT came out. This was nonsense, but it didn’t matter b/c it was (and to a large extent) a popular narrative that moved the stock market.

This genius is now a LinkedIn-recognized Top Voice on AI. It’s a crazy world. Image Source

This is bad enough by itself. Open Source Foundation Models make it much worse. Suddenly, Google has to contend against (phantom) enemies on two fronts. And unlike OpenAI/big Tech, Open Source is fluid, fast, and free. This makes it very good at eating market share against enterprise services. It’s also extremely unpredictable.

Even if the FMs don’t threaten Google’s business (and I repeat: neither Open Source nor OpenAI threatens Search/Ads), public perception has already pushed them against each other. So now Google has to watch out for it.

By throwing regulations around Open Source, they can choke it out. This is pre-emptive damage control. A few million to lobby and sell the apocalypse from Open Source is a small price to pay if it means bringing stability to the industry by keeping AI in the hands of slower-moving and predictable corporate entities.

All of this because Google has consistently failed to communicate how good they are. And that is why Google needs to fire whoever is responsible for the AI PR and hire the guys from Anthropic instead (Anthropic’s brilliant Media Strategy will be its own article, coming soon). Or holla at your boy, I’ll give your researchers the love they deserve.

Why MS and OpenAI love Regulations

Just as with Google, Open Source adds a lot of chaos to AI for MS and OpenAI. This is not good for MS, which has a lot of money invested into OpenAI. Worse still, the hype is dying down (one of my investor friends told me that AI was no longer moving investments as much as it did a few months ago). The monthly seed values for AI startups is dropping. Gen AI came into the bar with a fancy suit and a loud voice but wasn’t too good at maintaining attention. So he ends up going home alone. Tale as old as time.

If you look behind the curtain, many of GPT’s promises have not materialized. AI isn’t taking over most lucrative business opportunities. I’ve spoken to a lot of readers who are struggling to bring in good performance with Vector DBs+OpenAI Models for RAG. Free piece of advice for you- look into indexing data with FAISS (which was released in 2017). This works absurdly well (and in many ways inspired modern Vector DBs).

Data Labelling, Transcription, and Annotation are good use cases, but any large model can do them reasonably well (for most general cases). This is where MS has a commercial interest in stopping competition for OpenAI. It’s no coincidence that Sam Altman changed tunes on the need for oversight of AI after strong open-source models with commercial licenses were released. Tie up Open Source in regulation, and OpenAI can waltz in and do its thing.

This double incentive is why Microsoft has been the most aggressive player for regulatory capture. They have the most to gain from the whole thing. OpenAI’s great public image also means that once that regulation takes hold, it will be in the best place to build up hype. People don’t like Meta, Google has broken its own knees, and Amazon has been silent. In the court of public opinion, OpenAI always wins. Microsoft needs to kick back and let things happen.

To finish off 2021, I wrote an article where I said that Microsoft had put itself in the best position to dominate AI. My analysis was based on the fact that they had made great investments into all the major components of AI, and could thus play to adapt to every development. Watching them navigate AI is surreal. Nadella and the rest of the MS management have played their cards phenomenally. If I could, I would hang up their strats in the Louvre (I’d pick this over the Mona Lisa). It sucks that they’re going against Open Source, but they’re only playing the game (not condoning their lobbying and market manipulation but god damn bros, well played).

Google might be ahead in research, but it doesn’t matter b/c no one will hear about their work. MS can continue to dictate the narrative and move the needle in its favor. For publicly traded companies, that’s what really matters.

Let’s move on to companies that have been surprisingly quiet in this whole thing. Amazon and Apple have been uncharacteristically detached from the whole debate. Let’s cover why.

Why Amazon and Apple are just chilling

Aside from their investment into Anthropic, Amazon hasn’t hit the headlines. This is deliberate. Like Google, Amazon isn’t threatened by Foundation Models. Unlike Google, no one thinks they are. Amazon is in a position that Google is undoubtedly jealous of, free from public inspection and able to invest in its core capabilities without worry.

Apple is in a similar position ti Amazon. So they aren’t worried about Foundation Model debates and can stick to their game. They recently released a PyTorch-like ML library specialized for Apple Hardware. This is a minor endorsement for Open Source, but more importantly, gives people a stronger buy-in to the Apple Ecosystem.

Both these tech giants can continue to live how they please, without having to get involved in this contentious tug-of-war. They have the resources to catch up and pivot to however the cookie crumbles, so they just need to sit out and let things happen.

Nothing else to say here. Let’s move on to Meta, and why they have been such strong warriors for Open Source.

Why Meta Champions Open Source

To beat a dead horse- foundation models don’t threaten Meta either. But they will benefit from AI Developments. This means that for Meta, Open Source is a huge winner. They can outsource large chunks of R&D to open-source contributors, who will share things for free. They just need to fold these developments back into their products.

This is especially important for Meta because of the Metaverse (bet you weren’t expecting that). Metaverse is a moonshot project, one that requires a lot of setup to get going. A healthy open-source ecosystem will cut down R&D for them. A few million is a relatively small price to pay for this (+ you get brownie points and get more people into your ecosystem). This is also why so many big names joined Meta’s alliance: they benefit positively from the downstream effects of OSS; they get access to a deep network in the industry; and they get prestige points.

All in all, we can see various financial interests behind companies and their differing stances on the Regulation. All in all, we’re likely going to see the balance of power shift towards open source since more organizations have money to gain. In the meantime, politicians will try to push some bills for optical reasons or to gain control. If their regulation stifles open source, it’ll be our job to fight back. Learning to unpack the financial motives behind developments is key for that.

If you liked this article and wish to share it, please refer to the following guidelines.

That is it for this piece. I appreciate your time. As always, if you’re interested in working with me or checking out my other work, my links will be at the end of this email/post. And if you found value in this write-up, I would appreciate you sharing it with more people. It is word-of-mouth referrals like yours that help me grow.