With MAI-1, Is Microsoft Trying to Compete with OpenAI?

0:00Today on the AI Daily Brief, Microsoft is reportedly launching a model to compete with GPT4. Before that on the brief, is Stack Overflow's new deal with OpenAI the end of Stack Overflow? The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, check out the Discord linked in the show notes.

0:24Welcome back to the AI Daily Brief, kicking off with today's AI headline news. Yesterday, Stack Overflow and OpenAI announced a new partnership. On an announcement blog post, Stack Overflow wrote, Today the company's announced a new API partnership that will empower developers with the collective strengths of the world's leading knowledge platform for highly technical content with the world's most popular LLM models for AI development. So what does this partnership include? Well, one piece of it is that OpenAI will have access to Stack Overflow's Overflow API product and will, quote, collaborate with Stack Overflow to improve model performance for developers who use their products.

0:58The goal here is to help OpenAI improve its models using, quote, enhanced content and feedback from the Stack Overflow community while also providing attribution to that community. Indeed, they say OpenAI will surface validated technical knowledge from Stack Overflow directly into ChatGPT, giving users easy access to trusted, attributed, accurate, and highly technical knowledge and code backed by the millions of developers that have contributed to the Stack Overflow platform for 15 years. The other part of the partnership is that Stack Overflow will utilize OpenAI models as part of the development of their own Overflow AI.

1:27Now, this is a notable partnership for a couple of different reasons. Stack Overflow has been largely seen by many to be one of the chief victims of the AI era. The chart that's currently on the screen shows the decline of Stack Overflow's traffic going back all the way to the release of GitHub Copilot in October 21, but especially accelerating with the release of ChatGPT in November of 2022. Between a peak of around 20 million daily page views in the 2020-2021 era, down to around 14 million in the wake of the release of Copilot, and going all the way down to 10 million, or half of their previous traffic by July of 2023.

2:00Francesco writes, So it finally happened. Stack Overflow announced its partnership, surrender, with OpenAI. I have mixed feelings about that and haven't used Stack Overflow in a while, and it always seemed like a useful but toxic place, especially for beginners. Commenter Sabakisi writes, Partnership or they'll die. No other option on the table. I have barely used Stack Overflow since ChatGPT was released. While the feelings were mixed for some, they were decidedly less mixed for others. A UI programmer at Epic Games tweeted, Stack Overflow announced that they are partnering with OpenAI, so I tried to delete my highest-rated answers.

2:32I found that Stack Overflow does not let you delete questions that have accepted answers in many upvotes because it would, quote, remove knowledge from the community. So instead, I changed my highest-rated answers to a protest message. Within an hour, mods had changed the questions back and suspended my account for seven days. Ben goes on, I'm requesting that my questions and answers be permanently deleted under GDPR. It's just a reminder that anything you post on any of these platforms can and will be used for profit. It's just a matter of time until all your messages on Discord, Twitter, etc.

2:58are scraped, fed into a model, and sold back to you. Lucille Danilov retweeted Ben's message and wrote, Be good guy dev. Contribute to Stack Overflow for years. Suddenly, Stack Overflow decides to sell out to OpenAI and scrape your data for no compensation. Protest against it by changing your highest rated answer, get banned, and have your original answer forcibly reinstated. How is this legal? Nick Scraft writes, Everyone should immediately stop contributing to the Stack Overflow and its network. The human touch is what made it unique. Delete your profile from SO and all your answers. Freeloaders are making money out of human contributors.

3:29OpenAI, Microsoft, and Google together kill the open web. Thousands of independent blogs and forums are now nowhere in search engines or pushed back to page two to support their AI in partnerships with Reddit, Stack Overflow, and more. Many humans contributed to these sites hoping to build a knowledge base for humanity, but now greedy people like Sam Altman and OpenAI are taking over everything. I share these not because I think they're the right take, but just to show how intensely people feel about this. It is a complicated issue that is completely uncomplicated for some. I think it's an indicator of the portion of the developer community, and indeed probably the larger community, that feels like any form of AI training is just another word for theft.

4:02At the same time, however, this sort of deal with publishers, where it's not so much about OpenAI getting access to their data, but OpenAI actually attributing things to them as well, seems to be the way the winds are blowing. Right now, it's kind of split between newspapers, for example, who are suing OpenAI and newspapers who are cutting these deals. And I think that pretty well demonstrates the particular bifurcation of this moment. Now, speaking of OpenAI, the information is reporting that two senior OpenAI executives have left the company. Those include Vice President of People, Diane Yoon, and Head of Nonprofit and Strategic Initiatives, Chris Clark.

4:34I'm always a little skeptical of this type of story. Especially in the technology space, people move between companies all the time. Executives leaving a company doesn't necessarily mean that something bad is going on in that company. It could be for any number of personal reasons. It could be because these folks were able to cash out some chunk of what they had vested with that tender offer at an$86 billion valuation. Of course, what the press tends to notice is when there seems to start to be a pattern. The information, for example, connects us to OpenAI co-founder Andre Carpathy leaving earlier this year, and there are some other notable departures as well.

5:05But for now, it feels to me more like the natural ebbs and flows of people in and out of OpenAI rather than a smoke signal that something really bad is going on. Finally today, some announcements in the Apple and chip world. First, Apple has unveiled its latest iPad Pro, and it's the first time that we're seeing its new M4 chip. During the presentation, Tim Millett, the Vice President of Platform Architecture at Apple, said, With this level of performance, the neural engine in M4 is more powerful than any neural processing unit in any AI PC today. As Reuters writes, Debuting its latest chip in a tablet rather than its Mac laptops is unusual for Apple, and suggests it is eager to give app makers a head start creating AI-related software ahead of its annual Software Developer Conference next month.

5:46Continuing, they write, Precisely what AI features the new M4 chips could power might not become clear until June 10th at Apple's annual Worldwide Developers Conference, where it often shows new capabilities for Siri, its voice assistant, as well as the rest of its operating systems. Ben Bajoran, the chief executive of consulting firm Creative Strategies, said of the announcement, Essentially, there was a lot of foreshadowing on where they can go or will go with generative AI on their hardware. So we have this one official announcement, but then we also got a report from the Wall Street Journal that Apple has also been working on custom chips to run AI in data centers.

6:17According to reports, the server project is internally codenamed Project ACDC, which stands for Apple Chips in Data Center. The journal writes Project ACDC has been in the works for several years, and it is uncertain when the new chip will be unveiled, if ever. For now, it's mostly just being taken as yet another sign that Apple is indeed quietly working on AI stuff and that the market should price it as such. However, that is going to do it for the headline section of today's AI Daily Brief. Next up, the main part of the episode. Hello, AI friends. Today, I want to tell you about our platform, Super Intelligent.

6:50In short, it's a platform for useful, practical, immediately applicable AI learning. We have nearly 400 video tutorials, each of which comes with step-by-step how-tos. And the idea is to get you actually using these AI tools we talk about every day in a matter of minutes to actually solve problems, create new opportunities, and just do really cool things. To learn more and subscribe, go to besuper.ai. And if you do decide to subscribe, use code podcast for 50 % off your first month. Again, that's besuper.ai. Welcome back to the AI Daily Brief. For a while now, the AI industry has been closely watching the evolving relationship between Microsoft and OpenAI.

7:30There have been some big questions about whether Microsoft is over that relationship, just trying to hedge their bets, or going in fundamentally different directions. The latest saga in this story comes with reports about MAI 1. The information writes, Meet MAI 1, Microsoft Readies New AI Model to Compete with Google and OpenAI. So let's see what this model is, what the community's take on all this is, and what it means for this relationship in the long run. The information writes, For the first time since it invested more than$10 billion into OpenAI in exchange for the rights to reuse the startup's AI models, Microsoft is training a new in-house AI model large enough to compete with state-of-the-art models from Google, Anthropic, and OpenAI itself.

8:08Now, they go on to write that the model is internally being referred to as MAI-1 and is being overseen by Mustafa Suleiman. Now, that's notable because the last time that we were having this conversation about Microsoft and OpenAI was when Microsoft hired the majority of the startup Inflection, as well as paying$650 million for the rights to its IP. It was an acquisition in anything but name. Mustafa was, of course, not only the CEO and founder of Inflection, but before that, he had been one of the co-founders of DeepMind who sold that company to Google. It felt to some like Suleiman and Inflection were being hired into the role that had for a time during the whole Sam Altman saga been where Altman himself might be headed.

8:44Remember, when Altman was fired as CEO, for a moment it appeared that he would be heading to Microsoft to lead a new AI division there. Obviously, that's not how that situation played out, but it seems in retrospect like either A, Microsoft had already had a plan like that, or B, they liked the idea so much that even after Altman went back to OpenAI, they decided to go through with it as well. Now, as for MAI-1, the information says that this is not a model that's carried over from inflection, that it is instead a homegrown Microsoft model. According to sources, MAI1 will have roughly 500 billion parameters, as opposed to GPT-4's estimated more than 1 trillion parameters, but a lot more than other open-source models like Meta and Mistral's 70 billion parameters.

9:23The information's analysis is this. That means Microsoft is now pursuing a dual strategy of sorts in AI, aiming to develop both small language models that are inexpensive to build into apps and that could run on mobile devices alongside larger state-of-the-art AI models. The last note from the report is that Microsoft has been setting aside a large cluster of servers with NVIDIA GPUs, as well as compiling training data. So what do the community think? Well, Bindu Reddy from Abacus writes, As I predicted, Microsoft is training its own LLM. It's called MAI-1, a 500 billion parameter model, and may be previewed at their build conference.

9:54When this model becomes available, it will only be natural for Microsoft to push this instead of the GPT line. As predicted, OAI and Microsoft are becoming more and more competitive. Any which way, LLMs have become a commodity. Andrew Curran writes there have been indications for some time that Microsoft was intending to take their own path, and were planning to train a bespoke model to compete with GPT-4. We now have a name, MAI-1. Microsoft, for their part, tried to tamp down the idea that this was somehow representative of a change in their relationship with OpenAI. Kevin Scott, the CTO of Microsoft, wrote, I'm not sure why this is news, but just to summarize the obvious, we build big supercomputers to train AI models.

10:31Our partner OpenAI uses these supercomputers to train frontier-defining models, and then we both make these models available in products and services so that lots of people can benefit from them. We rather like this arrangement. We've been at it for almost five years now. Each supercomputer we build for OpenAI is a lot bigger than the one that preceded it, and each frontier model they train is a lot more powerful than its predecessors. We will continue to be on this path, building increasingly powerful supercomputers for OpenAI to train the models that will set the pace for the whole field well into the future.

10:57There's no insight to the increasing impact that our work together will have. We also, for years and years and years, have built AI models in Microsoft, in MSR, and in our product groups. AI models turn out to be interesting things to work on, and our researchers do great work studying and building them. AI models are used in almost every one of our products, services, and operating processes at Microsoft, and the teams making and operating things on occasion need to do their own custom work, whether that's training a model from scratch or fine-tuning a model that someone else has built. There will be more of this in the future, too.

11:24Some of these models have names like Turing and MAI. Some, like Fi, for instance, we even open source. I know the way I've said it isn't all that dramatic, but it is reality, and it's a plenty exciting reality for all us geeks given how hard all of this is to do in practice. So basically the argument here is that just because Microsoft is working on other models doesn't mean that it doesn't like OpenAI's models or its partnership with OpenAI. It just has other uses and other needs for building other models. There's also been some interesting stuff coming out recently about how the relationship got started, specifically the involvement of Bill Gates.

11:55For example, a recent article reads, In 2017, just before Microsoft forged a partnership with a then-relatively-unknown startup called OpenAI, Bill Gates shared a memo with CEO Satya Nadella and a small group of the company's top executives. A new world order, Gates predicted, would soon be brought on by what he called AI agents, digital personal assistants that could anticipate our every want and need. These agents would be far more powerful than Siri and Alexa with godlike knowledge and supernatural intuition. Gates wrote, agents are not only going to change how everyone interacts with computers, they're also going to upend the software industry, bringing about the biggest revolution in computing since we went from typing commands to tapping on icons.

12:29Business Insider went on, Publicly, Gates has been almost entirely out of the picture at Microsoft since 2021. In fact, Gates has been quietly orchestrating much of Microsoft's AI revolution from behind the scenes. Current and former executives say Gates remains intimately involved in the company's operations, advising on strategy, reviewing products, recruiting high-level executives, and nurturing Microsoft's crucial relationship with Sam Altman. One Microsoft executive said, What you read is not what's happening in reality. Satya and the entire senior leadership team lean on Gates very significantly.

12:57His opinion is sought every time we make a major change. Another email that recently came out was one from CTO Kevin Scott, who we just heard from. It was from June 12, 2019. It was addressed to Satya Nadella and Bill Gates and was called Thoughts on OpenAI. Kevin writes, The thing that's interesting about what OpenAI and DeepMind and Google Brain are doing is the scale of their ambition, and how that ambition is driving everything from data center design to compute silicon to networks and distributed systems architectures to numerical optimizers, compilers, programming frameworks, and the high-level abstractions that model developers have at their disposal.

13:27When all these programs were doing was competing with one another to see which RL system could achieve the most impressive game-playing stunt, I was highly dismissive of their efforts. That was a mistake. When they took all of the infrastructure that they had built to build NLP models that we couldn't easily replicate, I started to take things more seriously. And as I dug in to try to understand where all the capability gaps were between Google and us for model training, I got very, very worried. As Bloomberg summed up, Microsoft's motivation for investing heavily in partnering with OpenAI came from a sense of falling badly behind Google.

13:55Said Scott in the email, we are multiple years behind the competition in terms of machine learning scale. Satya Nadella endorsed Scott's email, forwarding it to Chief Financial Officer Amy Hood and saying it explains, quote, why I want us to do this. Daryl Bossenjo writes, both the competitive awareness to understand how far ahead Google was and the realization that Microsoft's Bing teams would not be able to compete without help are great. So of course the question becomes, have things changed so fundamentally in five years that Microsoft no longer feels behind? Semi-Analysis explored this in a piece today called OpenAI is doomed?

14:26E2 Microsoft? In it they discuss the changing landscape of compute resources, the new emergence of models from China that are cheaper to run and highly performant, and then there's a section they have called Is Microsoft Even Committed? They write, Microsoft is spending over$10 billion in CapEx directly for OpenAI, but they are not directing most of their GPU capacity to OpenAI. The majority of Microsoft's planned over$50 billion annual spend on AI data centers is going to internal workloads. Much of this has been inferenced to deploy OpenAI models in their own products and services, but that's changing.

14:55Microsoft is forced into looking for contingency plans because of OpenAI's bizarre structure. OpenAI is a non-profit whose primary goal is creating artificial general intelligence that is safe and benefits all of humanity. OpenAI can and will break the agreement that enables Microsoft to have access to OpenAI's model with zero recourse from Microsoft. And then they quote an OpenAI document that reads, While our partnership with Microsoft includes a multi-billion dollar investment, OpenAI remains an entirely independent company governed by the OpenAI non-profit. Microsoft is a non-voting board observer and has no control.

15:25AGI is explicitly carved out of all commercial and IP licensing agreements. The board determines when we've attained AGI. Again, by AGI we mean a highly autonomous system that outperforms humans at most economically valuable work. Such a system is excluded from IP licenses and other commercial terms with Microsoft, which only apply to pre-AGI technology. As Semi-Analysis continues, the most worrisome thing for Microsoft here is that OpenAI's board can decide at any moment, with no voting input whatsoever from Microsoft, that they have achieved AGI and Microsoft is not entitled to the IP that was created with their investment dollars.

15:54When you stack that on top of the massive governance issues OpenAI already has related to their non-profit and for-profit arms, Microsoft must put a contingency plan in place. The Semi-Analysis authors then draw the line between a bunch of internal initiatives leading all the way to MAI 1. They call it Microsoft's first big effort at hitting GPT-4 class and say that the goal is to have their own in-house from scratch GPT-4 class model by the end of this month. So what to make of all of this? With no knowledge whatsoever of what's actually going on behind the scenes, it seems to me like everything is kind of right.

16:24What I mean by that is that I don't think that Kevin Scott is putting on a show of saying that they continue to plan to have a great relationship with OpenAI for years to come. I think that relationship has been extremely profitable for them. I think they see a lot of potential in it. I think that they believe that OpenAI has some secret sauce that keeps them ahead of the curve. So I take all of what he said in that LinkedIn post at face value. At the same time, I also think that that semi-analysis analysis is correct. The simple fact of having this AGI clause in the deal terms means that Microsoft can't rely on that relationship forever.

16:55And what's more, they've seen how quickly and how seemingly capriciously the OpenAI board can act. It's not surprising then that even if they are not wishing for a breakup with OpenAI, they are hedging their bets more aggressively since the end of last year. Anyways, it is a fascinating time. It will continue to be a fascinating time. And I will be here to tell you all about it as it happens. For now, however, that is going to do it for the AI Daily Brief. Until next time, peace.

17:30Good night.