Tech Brew Ride Home - Tue. 04/02 – Amazon To Enter The AI Group Chat

Episode Date: April 2, 2024

Big trove of AT&T customer data dumped online. Microsoft is unbundling Teams. Amazon is readying its own big LLM. We should probably assume Section 230 does not cover AI. And what do you name an AI supercomputer? Stargate, apparently. Links: AT&T resets account passcodes after millions of customer records leak online (TechCrunch) Exclusive: Microsoft to separate Teams and Office globally amid antitrust scrutiny (Reuters) Amazon scrambles for its place in the AI race (The Verge) Gen-AI Search Engine Perplexity Has a Plan to Sell Ads (Adweek) The AI Industry Is Steaming Toward A Legal Iceberg (WSJ) Microsoft and OpenAI Plot $100 Billion Stargate AI Supercomputer (The Information) Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to the Tech meme right home for Tuesday, April 2nd, 2024. I'm Brian McCullough today. Big trove of AT&T customer data dumped online.

Microsoft is unbundling teams. Amazon is rating its own big LLM. We should probably assume Section 230 does not cover AI. And what do you name an AI supercomputer? Stargate, apparently. Here's what you miss today in the world of tech. Some big news we missed this weekend. A bunch of data was dumped on the dark web.

and AT&T was forced to confirm that it appears to be from 2019 or earlier. It's their data. It has personal data of 7.6 million current and 65.4 million former account holders. So if you got a notification to reset your pass codes on AT&T recently, this is why, quoting TechCrunch. In a statement provided Saturday, AT&T said AT&T has launched a robust investigation supported by internal and external cybersecurity experts, Based on our preliminary analysis, the dataset appears to be from 2019 or earlier, impacting approximately 7.6 million current AT&T account holders and approximately 65.4 million former account holders. AT&T does not have evidence of unauthorized access to its systems, resulting in exfiltration

of the dataset. The statement also said, TechCrunch held the publication of this story until AT&T could begin resetting customer account passcodes. AT&T also has a post on what customers can do to keep their account secure. AT&T customer account passcodes are typically four-digit numbers that are used as an additional layer of security when accessing a customer's account, such as calling AT&T customer service in retail stores and online. This is the first time that AT&T has acknowledged that the leaked data belongs to its customers, some three years after a hacker claimed the theft of 73 million AT&T customer records. AT&T had denied a breach of its systems, but the source of the leak remains inconclusive. AT&T said Saturday that, quote, it is not yet known whether the data in those

fields originated from AT&T or one of its vendors, end quote. In 2021, the hacker claiming the AT&T breach posted only a small sample of records, making it difficult to check if the data was authentic. Earlier in March, a data seller published the full 73 million alleged AT&T records online on a known cybercrime forum, allowing for a more detailed analysis of the leaked records. AT&T customers have since confirmed that their leaked account data is accurate. The leaked data includes AT&T, customer names, home addresses, phone numbers, dates of birth, and social security numbers, end quote. Microsoft plans to unbundle teams from Microsoft 365 and Office 365 globally amid antitrust scrutiny, after a similar unbundling in the EU in August of last year, quoting Reuters.

The European Commission has been investigating Microsoft's tying of office and teams since a 2020 complaint by Salesforce owned workspace messaging app Slack. Teams, which was added to Office 365 in 2017 for free, subsequently replaced Skype for business and became popular during the pandemic due in part to its video conferencing. Rivals, however, said packaging the products together gives Microsoft an unfair advantage. The company started selling the two products separately in the EU and Switzerland in October 1st last year. To ensure clarity for our customers, we are extending the steps we took last year to unbundled teams from M365 and O365 in the European Economic Area and Switzerland to customers globally. A Microsoft spokesperson said, doing so also addresses feedback from the

European Commission by providing multinational companies more flexibility when they went to standardize their purchasing across geographies. Microsoft said in a blog post that it was introducing a new lineup of commercial Microsoft 365 and Office 365 suites that do not include teams in regions outside the EEA and Switzerland and also a new standalone teams offering for enterprise customers in those regions. Starting April 1st, customers can either continue with their current licensing deal, renew, update, or switch to the new offers. For new commercial customers, prices for office without teams range from 775 to 575, depending on the product, while standalone teams will cost 525. The figures may vary by country and currency. The company did not

disclosed prices for current packaged products. Microsoft's unbundling may not be enough to stave off EU antitrust charges, which will likely be sent to the company in the coming months, as rivals criticize the level of the fees and the ability of their messaging services to function with office web applications in their own services, sources said, end quote. We've been talking about the big cloud players moving heavily into AI, Google, of course, with their own homegrown AI products, Microsoft, with Open AI largely, but Who's the biggest cloud player of them all? Amazon, right? Well, Amazon is investing in AI startups, as we've discussed, especially Anthropic, but wouldn't you expect them to want their own major

internal play as well? Sources say, Amazon's AGI team is aiming to outperform Anthropics' latest clawed models by the middle of this year using the company's forthcoming LLM, which is codenamed internally Olympus, quoting the verge. The logic is simple. Amazon needs to offer models through AWS that compete with the OpenAI-powered offerings of its Arch Cloud rival Microsoft, and Anthropic is the best alternative that exists. If we rewound the clock back to when Big Tech could make large acquisitions without them being blocked by regulators, I'm sure Amazon would have tried to buy Anthropic outright. Instead, it's passively investing billions of dollars and telegraphing that it only has a minority stake with no board seat. Conveniently,

for Amazon Anthropic has meanwhile agreed to spend $4 billion on AWS over the next several years. There are obvious parallels here to Microsoft's funding of Open AIs, ever-growing compute needs, but Amazon's relationship with Anthropic is far less cozy than it appears on the surface. In fact, another part of Amazon is trying to compete directly with Anthropics models. I've learned that Amazon's AGI team, led by SVP Roheet Prasad, has the aggressive goal of outperforming Anthropics' latest clawed models by the middle of this year. Its forthcoming flagship model internally codenamed Olympus is in training and quite large. large with hundreds of billions of parameters. As rumors swirl about OpenAI's GBT5 being on the near

horizon, time is of the essence. Amazon knows it is behind in the model race and its own employees are having difficulty waiting for Olympus 2. Anthropics models are far better and more performant than anything the company has made in-house to date, I'm told. And there are whispers of product teams across the company, including parts of AWS, switching to Anthropics models for the time being. Claude competes with us and makes us nervous, and Amazon Insider told me this week. For any team not training LLMs and only using them, Claude is the clear winner for now. I'm told there is no plan, however, to switch Amazon's AI shopping assistant Rufus to Claude from the standard Amazon-developed LLM that powers it now. When it's released later this year, Olympus will be plugged into nearly every part of Amazon and made available to other businesses through AWS.

Whether it will actually beat Anthropics models remains to be seen. It is the current pet project of CEO Andy Jassy, who made Prasad one of his direct reports last year when the AGI org was stood up. Amazon clearly sees frontier LLMs as core technology it needs to own, even if that potentially puts it on a collision course with its marquee AI investment, end quote. Quick follow-up to something we discussed on a bonus episode, AI search engine perplexity, whose website originally said that search should be, quote, free from the influence of advertising-driven models, now says they plan to sell ads. Don't call it a pivot, I guess, but quoting search engine land. How it will work. Related questions, which include links to sources, account for 40% of perplexity's

queries. So perplexity plans to add relevant, related brand-sponsored questions along with organic questions. What about brand safety? It will be a paramount priority. Dmitri Chevalenko, perplexity's chief business officer, told Ad Week, when will it come? While the details are few, all we know is it is coming in the upcoming quarters. On its about page, perplexity says, quote, Perplexity was founded on the belief that searching for information should be a straightforward, efficient experience free from the influence of advertising-driven models. However, Chevalenko said, quote, advertising was always part of how we were going to build a great business, end quote. In a separate recent interview with Wired, Perplexity founder and CEO Averind Serenovas discussed

ads saying that he wasn't against advertising. In fact, his vision for advertising, on the platform is to help advertisers understand who is searching and then bid on the most high-value traffic. Ads are not evil. When ads are done right, it's amazing, and generative AI is going to help us build even better targeting, he said, end quote. This is a stark contrast from OpenAI, CEO Sam Altman, who doesn't love ads. He has even called internet advertising a, quote, momentary industry. Also from the weekend, this Chris Mim's piece was interesting because it suggests Section 230 will not protect firms from lawsuits over the outputs of generative AI. Quoting Chris in the Journal.

If your company uses AI to produce content, make decisions, or influence the lives of others, it's likely you will be liable for whatever it does, especially when it makes a mistake. This also applies to big tech companies rolling out chat-based AIs to the public, including Google and Microsoft as well as well-funded startups like Anthropic and OpenAI. If in the coming years we wind up using AI the way most commentators expect, By leaning on it to outsource a lot of our content and judgment calls, I don't think companies will be able to escape from some form of liability, says Jane Bombower, a law professor at the University of Florida who has written about these issues. The implications of this are momentous. Every company that uses generative AI could be responsible under laws that govern liability for harmful speech and laws governing liability for defective products, since today's AIs are both creators of speech and products. Some legal experts say this may create a flood of lawsuits for companies of all sizes.

The legal logic is straightforward. Section 230 of the Communications Decency Act of 1996 has long protected internet platforms from being held liable for the things we say on them. In short, if you say something defamatory about your neighbor on Facebook, they can sue you, but not meta. This law was foundational to the development of the early internet and is arguably one reason that many of today's biggest tech companies grew in the U.S. and not elsewhere. But Section 230 doesn't cover speech that a company's AI generates, says Graham Ryan, a litigator at Jones Walker, who will be publishing a paper in the Harvard Journal of Law and Technology on the topic. Generative AI is the Wild West when it comes to legal risk for Internet technology companies, unlike any other time in the history of the Internet since its inception, he adds.

I spoke with several legal experts across the ideological spectrum, and none expect that Section 230 will protect companies from lawsuits over the outputs of generative AI, which now include not just text, but also images, music, and video, end quote. This is significantly less than $7 trillion, but a lot of money all the same. Sources are telling the information that Microsoft and OpenAI are planning a $100 billion data center project with a U.S.-based supercomputer dubbed Stargate to power OpenAI's products. quote, Microsoft would likely be responsible for financing the project, which would be a hundred times more costly than some of today's biggest data centers, demonstrating the enormous investment that may be needed to build computing capacity for AI in the coming years. Executives envisage the proposed US-based supercomputer, which they have referred to as Stargate, as the biggest of a series of installations the companies are looking to build over the

next six years. Microsoft's willingness to go ahead with the Stargate plan depends in part on OpenAI's ability to meaningfully improve the capabilities of its AI, one of these people said. OpenAI last year failed to deliver a new model it had promised to Microsoft showing how difficult the AI frontier can be to predict. Still, OpenAI CEO Sam Altman has said publicly that the main bottleneck holding up better AI is a lack of sufficient servers to develop it. If Stargate moves forward, it would produce orders of magnitude more computing power than what Microsoft currently supplies to OpenAI from data centers in Phoenix. And elsewhere, these people said, the proposed, the proposed the supercomputer would also require at least several gigawatts of power equivalent to what's needed

to run at least several large data centers today, according to two of the people. Much of the project cost would lie in procuring the chips, two of the people said, but acquiring enough energy sources to run it could also be a challenge. Such a project is, quote, absolutely required for artificial general intelligence. AI that can accomplish most of the computing tasks humans do, said Chris Sharp, chief technology officer of digital realty, a data center operator that hasn't been involved in Stargate. Though the project's scale seems unimaginable by today's standard, he said that by the time such a supercomputer is finished, the numbers won't seem as eye-popping. The executives have discussed launching Stargate as soon as 2028 and expanding it through

2030, possibly needing as much as 5 gigawatts of power by the end, the people involved in the discussion said. Altman and Microsoft employees have talked about these supercomputers in terms of five phases, with phase five being Stargate, named for a science fiction film in which scientists develop a device for traveling between galaxies. The codename originated with OpenAI, but isn't the official project code name that Microsoft is using, said one person who has been involved. The phase prior to Stargate would cost far less. Microsoft is working on a smaller phase four supercomputer for OpenAI that it aims to launch around 2026, according to two of the people. Executives have planned to build it in Mount Pleasant, Wisconsin, where the Wisconsin

Economic Development Corporation recently said Microsoft broke ground on a $1 billion data center expansion. The supercomputer and data center could eventually cost as much as $10 billion to complete, one of the people said that's many times more than the cost of existing data centers. Microsoft also has discussed using Nvidia-made AI chips for that project, said a person who has been involved in the conversations. Today, Microsoft and OpenAI are in the middle of phase three of the five-phase plan. Much of the cost of the next two phases will involve procuring the AI chips. Two data center practitioners who aren't involved in the project said it's common for AI server chips

to make up around half of the total initial cost of AI-focused data centers, other companies are currently building, end quote. Nothing for you today. Talk to you tomorrow.