Tech Brew Ride Home - Fri. 05/12 – Twitter Has A New CEO

Starting point is 00:00:00 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 TechMeme right home for Friday, May 12th, 2023. I'm Brian McCullough today. Elon says he's hired a new CEO for Twitter, and folks think they know who this person will be. The Claude Chatbot has expanded so much you can now write entire novels with it. Seeing what meta is doing with AI and ads makes me wonder about the future of all media. And of course, the weekend long read suggestions. Here's what you miss today in the world of tech. Elon Musk tweeted last night that he has hired a CEO for Twitter. really for X, as that's the legal name of the company now. But anyway, quote, she will be starting in around six weeks. Musk said he will become, quote, exec chair and CTO overseeing product, software, and CISOPs.

Starting point is 00:01:22 Quoting TechCrunch. Musk previously said he planned to step down as CEO and appoint a new chief executive before the end of 2023, noting that he would honor the results of a poll asking whether he should remain in charge of the company. However, he has previously not named any prospective candidates. Musk's decision to appoint a new Twitter CEO will likely appease Tesla investors, who have been concerned that Musk's time spent on Twitter is taking him away from his position as CEO of Tesla. Following the news that he found a replacement for the role, Tesla shares surged. Although Musk will soon no longer be CEO of Twitter, he still owns the company. Musk recently renamed Twitter Inc. to X-Corp.

Starting point is 00:02:01 He has long aspired to build what he. calls X, the everything app, end quote. So obviously, speculation immediately went rampant as to who this mystery woman might end up being, and speculation seems to have settled on. NBC Universal's head of advertising Linda Yakarino, quoting the Wall Street Journal and their sources. NBC Universal's head of advertising Linda Yakarino is in talks to become the new chief executive of Twitter, according to people familiar with the situation. Ms. Yakorino, chairman of Global advertising and partnerships at NBCU, has been with NBCU for more than a decade, where she has been an industry advocate for finding better ways to measure the effectiveness of advertising. As head of

Starting point is 00:02:44 NBCU's advertising sales, she was key in the launch of the company's ad-supported Peacock Streaming Service. Ms. Yaccarino would face immediate challenges as CEO, including wooing back advertisers who have typically provided the bulk of Twitter's revenue. In 2021, the year before Mr. Musk took Twitter private, advertising accounted for nearly 90% of the company's revenue. Twitter's revenue and adjusted earnings dropped by approximately 40% year over year in December, the Wall Street Journal previously reported. Mr. Musk said in April that Twitter would be cash flow positive as soon as this quarter. Of the top 100 advertisers on Twitter before Mr. Musk bought the company 37 spent nothing on Twitter advertising during the first quarter of this year, according to

Starting point is 00:03:25 Market Intelligence firm Censor Tower, while an additional 24 brands reduced their average monthly Twitter ad spending by 80% or more. Ms. Yacarino, who oversees roughly 13 billion in annual ad revenue, is well known for her tight relationship with marketers and ad agencies. Ms. Yaccarino has a reputation for hard-nosed negotiating tactics, and media buyers have described her as the Velvet Hammer. Last month, Ms. Yacarino conducted an on-stage interview with Mr. Musk at a conference in Miami Beach, Florida, during which she asked about his vision for Twitter and pressed him on his own sometimes controversial tweets, end quote. Anthropic has expanded Claude's context window from 9,000 to 100,000 tokens, or around

Starting point is 00:04:14 75,000 words which Claude can digest and analyze at one time. OpenAI's GPT4 has a context window of around 32,000 tokens, so basically this now means that with Claude, you can gin up a whole Harry Potter-sized book, not just the check. of a book in one go, quoting TechCrunch. Historically, and even today, poor memory has been an impediment to the usefulness of text-generating AI. As a recent piece in the Atlantic, aptly puts it, even sophisticated generative text AI like ChatGTPT has the memory of a goldfish. Each time the model generates a response, it takes into account only a very limited amount of text, preventing it from, say, summarizing a book or reviewing a major coding project.

Starting point is 00:04:59 But Anthropics trying to change that. Today, the AI research started. startup announced that it's expanded the context window for Claude, its flagship text-generating AI model still in preview, from 9,000 tokens to 100,000 tokens. Context windows refer to the text of the model considers before generating additional text, while tokens represent raw text, e.g. the word fantastic would be split into the tokens fan, tas, and tick. So what's the significance exactly? Well, as alluded to earlier, models with small context windows tend to forget the content of even very recent conversations leading them to veer off topic. After a few thousand words or so, they also forget their initial instructions, instead extrapolating their behavior from the

Starting point is 00:05:41 last information within their context window rather than from the original request. Given the benefits of large context windows, it's not surprising that figuring out ways to expand them has become a major focus of AI labs like OpenAI, which devoted an entire team to the issue. OpenAI's GPT4 held the previous crown in terms of context window sizes weighing in at 32,000 tokens on the high end, but the improved Claude API blows past that. With a bigger memory, Claude should be able to converse relatively coherently for hours, several days even, as opposed to minutes, and perhaps more importantly, it should be less likely to go off the rails. In a blog post, Anthropic touts the other benefits of Clause's increased context window, including the ability for

Starting point is 00:06:25 the model to digest and analyze hundreds of pages of material. Beyond reading long text, the upgraded Claude can help retrieve information from multiple documents or even a book, Anthropic says, answering questions that require synthesis of knowledge across many parts of the text. Anthropic lists a few possible use cases like digesting, summarizing and explaining documents such as financial statements or research papers, analyzing risks and opportunities for a company based on its annual reports, assessing the pros and cons of a piece of legislation, identifying risks, themes, and different forms of argument across legal documents, reading through hundreds of pages. of developer documentation and surfacing answers to technical questions, and rapidly prototyping

Starting point is 00:07:06 by dropping an entire codebase into the context and intelligently building on or modifying it. The average person can read 100,000 tokens of text in around five hours, and then they might need substantially longer to digest, remember and analyze that information, Anthropic continues. Claude can now do this in less than a minute. For example, we loaded the entire text of the Great Gatsby into Claude and modified one line to say, Mr. Carraway was a soft engineer that works on machine learning tooling at Anthropic. When we asked the model to spot what was different, it responded with the correct answer in 22 seconds, end quote. In an embargoed press event yesterday, I was present for Meta's announcement of AI Sandbox, a package of generative

Starting point is 00:07:53 AI tools to help select advertisers create text variations, background images from text, and cropped images for their ads used on meta's platforms, coding TechCrunch. The first feature lets brands generate different variations of the same copy for different audiences while trying to keep the core message of the ad similar. The background generation feature makes it easier to create different assets for a campaign. Finally, the image cropping feature helps companies create visuals in different aspect ratios for various mediums, such as social posts, stories, or short videos like Reels. The company said that these features are available to select advertisers at the moment, and it will expand access to more advertisers in July. While Meta is releasing some lightweight

Starting point is 00:08:34 generate generative AI features for advertisers, some ad tech startups are heavily leaning into it. Omniki, which presented at TechCrunch Disrupt last year, used OpenAIs, Dolly 2 and GPT3 to create ads. Movio, which counts IDG, Sequoia Capital China, and Baidu Ventures as its backers, is using Generative AI to create marketing videos as well, end quote. As I say, Meta invited me to attend this announce slash demo yesterday morning, and as I sat there seeing what these tools could do, I, was thinking, of course, that this was the obvious application for AI in this context. I bet it will be pretty powerful. But also, man, if you need another reminder, let me tell you that the monoculture

Starting point is 00:09:15 is over. Let me tell you what I mean. You know how we don't all listen to the same top 40 songs on the radio anymore? We all have our own playlist. And we don't watch the same TV shows. The already fractured cable TV landscape is now programming that we choose, not what is delivered to us by a fixed window of channels. One of the last things to my mind that we all kind of share, that we all kind of experienced together identically, was ads. I know that rotating ads are optimizing to send a specific version of an ad to a specific person is not new, but in general, even if the ad you saw was slightly different than mine, or if I never saw the ad you saw because I'm not the target market advertisers, to this point have only had a finite set of ad variations for lots of practical

Starting point is 00:10:02 and, you know, money reasons? Well, no more. What I saw yesterday presages a world where no one ever, ever sees the same ad as someone else, or even the same ad twice, maybe. Each ad could eventually be iteratively tweaked on the fly just for you or me for one specific moment. Instead of there being 50, 20, or 60 versions of an ad, there will now be infinite versions. They'll be as unique as fingerprints. So now, extrapolate out from there. What if, in the future, all of media is like this. You or I will never hear the same song because it is iteratively created on the fly just to please whoever is listening. We just spoke about that Google AI music generator this week. But what if also, no two people ever see the same movie because each movie is iteratively generated to the exact

Starting point is 00:10:56 characters, scenes, or plot that is designed to please one specific viewer at a time. Like, what I'm kind of saying is on a far enough along timeline, forget about creating a piece of content that will be liked by the largest amount of people, and then you can sell that to the largest amount of people. You just have one machine that creates content for everyone at any moment, at any time tailored specifically to them. It's not a theme. It's not individual. It could be, you know, what you feel like being amused by at the time. We're approaching Soma level here, people. Time for the weekend long-reed suggestions. Let's keep going with AI speculation.

Starting point is 00:11:44 Now that lawyers can use GPT4 to do legal research, now that you could load entire documents into Clawed, you could draft documents, analyze contracts. Is this the end of the billable hour? Quoting the journal. A March report by Goldman Sachs predicted that 44% of legal work can be automated using emerging AI tools. The same month, a paper by researchers at Princeton University, the University of Pennsylvania and New York University found that the industry's most exposed to occupational change from generative AI were legal services and securities, commodities, and investments. A lawyer's brain is basically a massive database of cases and precedence, said Min Q. Jung, a former practicing attorney who co-founded a firm called Latch, which uses

Starting point is 00:12:27 GPT4 to simplify the contract review and redlining process for lawyers. It's something a computer can do much more effectively than a human could, end quote. Latch launched in early April and already has a waitlist of more than 80 companies, including law firms and in-house counsel, Mr. Jung said. Global firm Allen and Overy said thousands of lawyers are now using another tool, using GPT4, called Harvey for tasks such as legal research, drafting documents, and contract analysis. The firm's attorneys report spending less time locating hard-to-fine case law, completing analyses and answering questions clearly and succinctly. Harvey hasn't replaced the work of lawyers, but instead provides a head start, they said.

Starting point is 00:13:09 One partner described the impact as, quote, having an extra junior resource available to you at any time of day. Will it mean fewer billable hours, the basis of a law firm's income stream? Yes, it's a possibility, said David Lucking, a partner at the firm. still the adoption of AI at Allen Overy doesn't necessarily mean there would be less need for a human element, he said. Then, given last week's bonus episode, I also wanted to share an oral history of BuzzFeed News from BuzzFeed News. I won't quote from it because that kind of doesn't work for oral histories, but check that out if you want more in-depth behind the scenes about what we talked to Ben Smith about last week. Then from the New York Times, we know the world is worried about

Starting point is 00:13:50 all of the silicon productions sort of getting bottlenecked in Taiwan, but what if Taiwan itself is facing a key bottleneck, which is talent demographic issues? Today, many at the top of Taiwan's semiconductor industry fear that tiny island territory will not be able to sustain the growing demand for a new generation of engineers. A shrinking population, demanding work culture, and an abundance of competing tech jobs have meant workers have become ever more scarce. Taiwan's talent crisis is intertwined with TSM's success. The company's employee count has grown almost 70% over the past decade, while Taiwan's birth rate has plummeted by half. Startups in promising fields like artificial intelligence are

Starting point is 00:14:32 luring top engineers. In recruiting, TSM must compete with the internet companies like Google and foreign semiconductor companies like ASML of the Netherlands, which generally offer better work-life balance and perks like free food. The challenges facing Taiwan's ship industry come amid a global crunch. In China, where officials have sought to lure Taiwanese engineers to build up its fledgling chip industry, the state-backed Chinese Academy of Sciences, has fretted about a, quote, serious shortage of qualified workers. By one estimate, China's microchip industry was short 200,000 people, end quote. And finally today, just a great sort of spy-clime story from Businessweek, the story of the woman who tried to steal the secret of Coca-Cola.

Starting point is 00:15:15 Not the secret you're thinking of, not the formula for Coke itself, the other big secret quote. Anytime a company lays someone off, there's a possibility the person will take something with them. Coke, holder of the world's most famous trade secret, was particularly attuned to that risk. It had an intelligence bureau-style classification scheme, like other corporations that deal in proprietary information, and it had software that tracked employees' data use. That summer, as more and more employees learned they were leaving, the data loss prevention system began to ripple with alerts. To say that the activity blew up in the DLP system would be a bit of an understatement. A Coke Information Security Manager later testified. Much of that activity resulted from employees reclaiming personal files they'd stored on

Starting point is 00:15:55 their work computers, tax returns, kids' school projects, bank loan information, but not all of it did. Shannon Yu in particular had access to some of the most closely held information at the company, a set of detailed chemical recipes for the two micron-thick plastic liners inside the beverage cans Coke filled and sold. A federal prosecutor would later describe these as the company's other secret formulas. Developed at great expense, they were likely even more important than the theatrically guarded recipe for Coke's namesake soft drink. That sugary acidic brew would, without a liner, devour the metal of its can. Editor's note makes you wonder if drinking Coke is worth it, if it can cut through metal like blood from an alien.

Starting point is 00:16:40 anyway. The liner formulas didn't actually belong to Coke, but to the multinational paint and coatings companies that were its partners. You was responsible for evaluating the formulas. She was one of only two people at Coke with access to many of the specifics. A month later, you would fly to Beijing to stand for another application, this time to a national grant program called the Thousand Talents. The money she wrote in her application would help the company she was co-founding build the first B-P-A-N-I coding production line in China. breaking the, quote, international monopoly in the global food container coatings industry. The files from her Coke computer were central to the plan, and she apparently was unaware of the

Starting point is 00:17:20 legal jeopardy that put her in. I'm the one taking all the risks in the end. She complained in Mandarin to one of her fellow aspiring co-founders on a we chat voice message. If anything happens to me, the money I've made wouldn't even be enough for the lawyer's fee, end quote. All right, it's another two bonus episode weekend this weekend. First up, on Saturday, my friend Lane Noonie has a new book out, where they make the, in my mind, correct historical assertion that the most important computer in the history of Apple as a company was not the Macintosh, but actually the Apple 2. So, tech history heads, get ready for an in-depth discussion of the first computer I ever used, the last computer that was truly a partnership between Steve Wozniak

Starting point is 00:18:08 and Steve Jobs, and quite obviously the computer that was Woz's masterpiece. Then I'm going to pull another nugget out of the internet history podcast that you might have missed, an interview with a co-founder of Tesla. What's this, Brian? You interviewed Elon Musk? No, it's complicated, but get ready for a long conversation with the Tesla founder who was there before Elon showed up. That's on Sunday. In the meantime, time, time to go pick up the kiddos from school and head for high rules. See you on Monday.

Tech Brew Ride Home - Fri. 05/12 – Twitter Has A New CEO

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.