AI M&A: Billion Dollar Acquisitions Reshape AI Landscape

0:00Today on the AI Breakdown Brief, we're looking at a set of acquisitions which could reshape the AI landscape. 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. Hello, friends. Welcome back to another AI Breakdown Brief. As I told you yesterday, this week, because I am traveling, I am just bringing you the briefs. There are no main episodes to follow. And the goal here is just to make sure that when it comes to the breaking headline news that I know a lot of you turn to the AI Breakdown for, we are still delivering on that promise.

0:35Now today we have a lot of really cool stories, so I'm going to dive right in. But if you're enjoying the AI Breakdown, I would ask one thing. If you haven't yet, go subscribe to the AI Breakdown newsletter. You can find it at the AIbreakdown.beehive.com. Now beehive is spelled B-E-E-H-I-I-V.com. Each morning I send out the five most important or interesting stories that I've seen in the past day, and I think it's a really great way to start your day in terms of what's going on in the AI space. I would so appreciate it if you would check it out and subscribe. It's totally free and a really great way to support the AI breakdown.

1:08All right, with that, let's turn to the AI breakdown brief. Today on the AI breakdown brief, billion-dollar acquisitions are reshaping the AI landscape. Welcome back to the AI breakdown brief, all the AI headline news you need in five minutes or less. Today, the big theme is absolutely huge, huge AI acquisitions. The biggest and the one that's getting the most buzz is that Databricks has signed a binding agreement to acquire Mosaic ML for$1.3 billion. Databricks calls themselves a data lake house vendor, meaning they combine data lakes, which is lots and lots of unstructured data, with data warehouses, which is lots and lots of structured data.

1:46And with their acquisition of Mosaic, they're looking to bring LLMs to their enterprise customers. Ali Godzi, the co-founder and CEO of Databricks, said, Every organization should be able to benefit from the AI revolution, with more control over how their data is used. Databricks and Mosaic ML have an incredible opportunity to democratize AI and make the lake house the best place to build generative AI and LLMs. In a tweet thread, he further said, Since November of last year, every customer I meet with asks me, How do I train and tune my own models? How do I keep my own data and own the IP? When we asked around about the companies at the forefront of this, everyone said Mosaic ML.

2:20So this is obviously a major trend right now. The first generation of LLMs in this new post-ChatGPT phase have obviously been general-purpose LLMs that are used by lots of different people for lots of different purposes. However, enterprises have real concerns and questions around the safety and security of their proprietary data. And as much as OpenAI is promising a new ChatGPT business version, for many enterprise customers, they're just going to want to train their own models. The Databricks-Mosaic-ML combination could be a very powerful offering in that context. Now that doesn't mean everyone is universally excited about this.

2:51Alex Valaitis writes this will have massive implications for the AI landscape, and says Mosaic was open source. It's unclear whether or not it will remain this way after the acquisition. If Mosaic ML were to become closed sourced, it would be a huge blow for the open source AI movement. Adding some weight to that argument, just earlier this week, Mosaic ML made news because its latest model had outperformed GPT-3, even though it was trained on only 30 billion parameters. GPT-3, by contrast, was trained on over 175 billion. As AI News puts it, this makes the Mosaic ML model more accessible to run on local hardware and significantly cheaper to deploy.

3:25Now, Alex also again points out that this creates a real significant battle for supremacy over the foundation model layer. As he writes, it sets up a battle between SnowflakeDB, Databricks, and OpenAI. Currently, it's unclear who will have long-term dominance over the LLM layer. Now, that was not the only big AI acquisition this week. Another one comes from fintech automation platform Ramp, who has acquired customer support AI company Cohere.io. Now, whereas the Mosaic ML acquisition shows a battle for supremacy on a key foundation LLM layer, Ramp's Cohere acquisition shows what's likely to happen a lot over the next couple years as industry-leading products turn to AI startups to integrate AI into their platforms more quickly than they could if they just built it on their own.

4:08A third major AI acquisition is that Thompson Reuters has agreed to acquire legal AI firm CaseTex for$650 million in cash. CaseTex builds products for the legal industry, such as CoCounsel, which is an AI legal assistant that's powered by GPT-4, which helps with things like document review, legal research memos, deposition preparation, contractual analysis, and more. Earlier this year, Thomson Reuters' chief financial officer had said that the company planned to spend$100 million this year to invest in AI, which would be separate from an M &A budget over the next few years of around$10 billion.

4:39I have a feeling that in the coming months, we're going to be able to talk about some AI acquisition somewhere nearly every day. Moving over to a new awareness survey from The Verge, the publication polled 2 ,000 people about how they use AI, what they think about it, what scares them about it, and more, and found some really interesting data. First, when it comes to who's heard of or used AI, ChatGPT unsurprisingly tops the list, with 57 % having used it or heard of it, and only 43 % saying they didn't know what it was. Bing Chat from Microsoft is at 46 % having used it or heard of it. Snap's MyAI is at 45 % showing the dominance of Gen Z, as we'll see in just a minute.

5:16And interestingly, MyAI from Snap is actually ahead of Google's BARD, which only 38 % of people have used or heard of. When it comes to the text-to-image generators, MidJourney is slightly ahead of Stable Diffusion, with 25 % of people having heard of or used MidJourney versus 23 % of people having used or heard of Stable Diffusion. Now, when it comes to which generations have interacted with this technology, these results look a lot more like what I would have expected than the results we shared last week of a different study. Millennials and Gen Z were clearly the most engaged, with 36 million millennials and 34.9 million Gen Z members having used AI versus just 4.8 million boomers and 15.8 million Gen Xers.

5:52When it comes to whether people think these technologies will have a big impact on society, the answer is a definitive yes. 74 % of people said that AI would have a big impact on society. That's versus 69 % for electric vehicles, 60 % for virtual reality, 52 % for augmented reality, and just 34 % for NFTs. Womp womp. Now, one of the most interesting bits of information was how people are using it. And the reason I thought this was interesting is just how diverse it is. 68 % of people who had used AI said they use it to answer a question. That was decidedly the most common use case. After that was brainstorming, a use case that I've talked a lot about here, with 54 % of people having used AI to brainstorm.

6:29After that, there is a ton of use cases that have between 25 % and 35 % of people who have used AI having used it for that. Data analysis is at 26%, writing stories is at 29%, artwork is at 27%, emails is at 25%, photo editing is at 37%, design is at 29%, coding is just at 18%, but I think that that probably just reflects that there are fewer coders in general. Another really interesting area is that there is a lot of very common sense agreement around things when it comes to AI regulation. 76 % of people said that there needed to be new laws and regulations around AI. 76 % said that AI should be required to be trained on fact-checked data sets.

7:0578 % said that AI-created digital content needs to state that it was created with AI. And 76 % said that video and audio deepfakes should be illegal if they're created without the real person's consent. Now, broadly speaking, when it comes to how excited or anxious people are about AI, like many of these polls, it's both. That was the most common answer with 32 % of people saying they felt both excited and anxious. 18 % said neither, but then the pure anxious beat out the pure excited by just a little bit. 29 % of people were anxious while 21 % of people were excited. Anyways, there's a lot more great info in there.

7:35I will include a link in the show notes so you can go check it out for yourself. Moving over to another trend in the LLM space, we talked before about how enterprises are looking for LLMs that can preserve and keep safe their proprietary data, but other contexts have slightly different trade-offs when it comes to what they need out of their large language models. For example, when it comes to education, if LLMs are to be really incorporated in standard practices, people have to be sure that they're not just going to hallucinate or make up information. MerlinMind has just announced that they are releasing three LLMs that are specifically designed for trustworthy generative AI in education.

8:08The announcement tweet says, the platform will enable teachers and students to have a generative AI experience that retrieves content from curriculum chosen by the user, not from the entirety of the internet. The result is an engagement that is curriculum aligned, hallucination resistant, and age appropriate. This reminds me of how I've talked about how I use ChatGPT a lot, using plugins like Xpapers to make sure that it's pulling information specifically from a source that I've chosen. Now, going back to that Verge study for a moment, for those 29 % of people who are anxious about AI, I'm not sure that this next piece of news will make them feel any better.

8:39Independent developer Stephen Sturges has announced that he is releasing something called Waldo 2.0. As he describes it, it's open source detection AI for your drone, satellite, or flying thing, and basically it does exactly what it sounds like. As a drone is flying over a scene, it can identify objects such as cars, houses, etc. in real time. Now of course, on the one hand, it's incredibly impressive technology that has a lot of really valuable use cases. Think search and rescue, archaeology, city planning. There's no shortage of really legitimate, above-board, non-horrifying surveillance use cases.

9:09But there are also a lot of Skynet horrifying surveillance use cases, and this thing is going to be open source, which again brings up the question of open source technology and why it's a little bit more complicated for some people than it seems like it might be in other software contexts. Still, it's worth checking out as it is an impressive piece of technology, if nothing else. Lastly today, speaking of an impressive piece of technology, a new documentary is about to come out that tells the story of a team that is working with AI to try to crack the code of whale songs. Yes, it's using AI to try to effectively interpret and translate whale calls into language that we can recognize.

9:43If you, like me, had about a dozen whale song tapes on your shelf in the early 90s, you're going to be pretty excited about this one. Anyways, guys, that is it for today's AI Breakdown Brief. Go check out the podcast and the newsletter version. Until next time, peace.