Tech Brew Ride Home - Tue. 02/25 – Claude 3.7 Sonnet

Episode Date: February 25, 2025

Anthropic releases its latest cutting-edge model, Claude 3.7 Sonnet. Would you tolerate ads inside your Excel spreadsheet if you could use Excel for free? Why some journalists are joining AI companies. And let me introduce you to “tiny teams” the new startup meme in Silicon Valley. Links: Anthropic launches a new AI model that ‘thinks’ as long as you want (TechCrunch) AI Startup Anthropic Finalizing $3.5 Billion Funding Round (WSJ) Microsoft Quietly Launched a Free Ad-Supported Office App, and No One Noticed (Beebom) Apple’s $500 Billion U.S. Investment Is Mostly Already in the Books (WSJ) Meet the journalists training AI models for Meta and OpenAI (NeimanLab) A.I. Is Changing How Silicon Valley Builds Start-Ups (NYTimes) 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, February 25th, 2025. I'm Brian McCullough today. Anthropic releases its latest cutting edge model, Claude 3.7 Sonnet. Would you tolerate ads inside your Excel spreadsheet if you could use Excel for free? Why some journalists are joining AI companies and let me introduce you to Tiny Teams, the new startup meme in Silicon Valley. Here's what you missed today in the world of tech. Anthropic has released Claude 3.7 Sonnet, a hybrid model that can produce fast responses or extended. step-by-step thinking and Claude, an agentic coding tool.

Here's everything you need to know, quoting TechCrunch. Anthropic calls Claude 3.7 Sonnet the industry's first hybrid AI reasoning model because it's a single model that can give both real-time answers and more considered thought-out answers to questions. Users can choose whether to activate the AI model's reasoning abilities, which prompt Cloud 3.7 Sonnet to think for a short or long period of time. The model represents Anthropics' broader effort to simplify the user experience around its AI products. Most AI chatbots today have a daunting model picker that forces users to choose

from several different options that vary in cost and capability. Labs like Anthropic would rather you not have to think about it. Ideally, one model does all the work. Cloud 3.7 Sonnet is rolling out to all users and developers on Monday, Anthropics said, but only users paying for Anthropics' premium cloud chatbot plans will get access to the model's reasoning features. Free Claude users will get the standard non-reasoning version of Claude 3.7 Sonnet, which Anthropic claims outperforms its previous frontier AI model, Cloud 3.5 Sonnet. Yes, the company skipped a number, end quote. And quoting the verge. The model costs the same to run as its predecessor, 3.5 Sonnet at $3 per million token inputs and $15 per million output tokens. While Claude still lacks real-time

web search like other models version 3.7's knowledge cutoff date of October 2024 is more up to date. Anthropic is also allowing developers to help. steer how the model thinks via its scratch pad and even dictate exactly how long it takes to respond. In addition to a new model, Anthropic is also releasing a limited research preview of its agentic coding tool called Claude Code. While Anthropic already powers AI coding tools like cursor, it's pitching Claude Code as an active collaborator that can search and read code, edit files, write and run tests, commit and push code to GitHub, and use command line tools, end quote. And quoting Bloomberg. If a user asks Claude 3.5 Sonnet to spend additional

time computing a response, the model will show written details of the chain of thought process. It follows a design choice that OpenAI, XAI, and Deepseek have also recently embraced. Anthropic also indicated that it remains focused on making so-called agents or software that can carry out more difficult tasks with little human oversight. The company said, Claude 3.7 Sonnet will include computer use, a tool the company first rolled out last year that can interpret what a user is seeing on their computer and with their permission take actions on their behalf, like browsing the internet, typing, and tapping buttons, end quote. By the by, since we're on the topic, Anthropic is reportedly finalizing a three and a half

billion dollar funding round, valuing it at $61.5 billion up from an $18 billion valuation last year. Its annualized revenue recently hit around $1.2 billion, according to sources, quoting the journal. Investors in the latest round include the venture firms, lightspeed venture partners, general catalyst, and Bessemer venture partners, the people said. Abidabie-Based investment firm, MGX is also in talks to participate. Anthropic trails market leader OpenAI and among consumer users, though its Claude Chatbot has become popular among programmers and business clients. The startup's annualized revenue and extrapolation of the next 12 months revenue based on recent sales, recently hit about $1.2 billion. One of the people said it is still losing money and will use

cash from the latest funding to support its efforts to train more powerful AI models. OpenAI told investors in an October funding round that it expected to generate $3.7 billion in revenue, last year, end quote. Yeah, Anthropic is in this weird spot. I'm not the first to point this out, but while the cool kids seem to love Anthropic and I use Claude myself, Open AI has like a hundred times the brand recognition among Normies, which matters in enterprise sales. It's sort of like the old nobody gets fired for buying IBM adage. Then, you know, Google has all the money in the world and is known to enterprises too. So what can Anthropic do to stand out? Here's a real wild shower thought for you. Remember how Chris and I were talking about the need for somebody to test the

M&A waters? Who is Anthropics' biggest backer to date, Amazon? I'm not saying this would happen, but what would be a wild test of the waters would be if Amazon tried to acquire Anthropic? Would you use a version of Microsoft Office that was free, but ad-supported? You might get the chance because Microsoft has quietly launched free, ad-supported desktop versions of Word, Excel, and PowerPoint for Windows with banner and video ads and limited functionality. Quoting B-Bomb. In all three apps, Microsoft Word, Excel, and PowerPoint, you will see a persistent ad banner on the right side. You can't remove it unless you buy the premium subscription. Moreover, a 15-second video ad will play muted in the office app every few hours.

Apart from that, you can't save documents locally. You will have to use OneDrive to save files. Of course, you can open the OneDrive app on your PC or go to its website to download these docs easily. Free users get 5 gigabytes of OneDrive storage. Moreover, most of the advanced features are not available on the ad-supported version of Microsoft Office. Notably, you can't install add-ons, add a watermark, analyze data, etc., end quote. Let me underline again. This is only available on Windows at the moment, and Microsoft told PC,

quote, Microsoft has been conducting some limited testing. Currently, there are no plans to launch a free ad-supported version of Microsoft Office desktop apps, end quote. But quoting Windows Central, to access the free version of Office, just skip the prompt to sign in when you first run an Office app. From there, you will be given the choice to continue to use Office for free in exchange for ads and limited features. In this mode, you can open view and even edit documents just like you can with the web

version of Office. Unfortunately, it looks like this free version of Office is. still in limited testing, as we were unable to skip the sign-in prompt when running office on our test machines. It's possible that Microsoft is currently only testing this version of office in specific regions, or is AB testing it with a small group of people first. The company hasn't officially announced this version of Office yet, nor does it acknowledge its existence in support documents, end quote. I'm old enough to remember 20 years or more ago when people floated the idea of ads supported Windows and Office, and people laugh that off.

one's laughing now, I guess. The Wall Street Journal ran the numbers on Apple's more than $500 billion spending announcement yesterday, and their analysis is basically this was maybe what they would have done anyway. Quote, unclear, though, is how much of the planned spending is actually new. Apple has spent about $1.1 trillion over the past four fiscal years on total operating expenses and capital expenditures, and Wall Street expects nearly $1.3 trillion in total spending over the next four years, according to consensus estimates by visible alpha. While Apple doesn't break out its expenses per geography, about 43% of its revenue comes from the America's region, which it defines as North and South America. Assuming the U.S. constitutes the large bulk of that number,

and if spending is about in line with revenue, then a rough figure of 40% of projected global spending through the 2028 fiscal year equates to about $505 billion. In short, Apple's announced figure is in line with what one might expect the company to be spending. anyway, given its financials. That isn't to say there won't be some new incremental spending domestically. The conditions are actually ripe for such a move, given Apple's continuing effort to diversify its manufacturing base outside of China. There is also domestic politics to consider, no small matter for a U.S. Consumer Electronics company that still builds the bulk of its products overseas. Indeed, the announcement seems to already have paid off. Thank you, Tim Cook and Apple,

President Trump exclaimed on his truth social platform Monday morning. And as big as Apple is now, even the world's most valuable company might be hard-pressed to come up with $500 billion in new funds to invest. In a report Monday, UBS analyst David Vote noted that the $95 billion in stock buybacks Apple made in its fiscal year that ended in September, consumed about 80% of the company's cash from operations. Therefore, Apple would need to increase balance sheet, leverage materially or reduce the buyback cadence to options we find highly unlikely, Voked, wrote. A huge jump in spending could take some shine off Apple's stock, another reason to suspect that what Apple's actually planning is something less dramatic. The stock has benefited in part from the company staying out of the AI spending race that has consumed Mega Cap Tech Peers, Microsoft, Amazon, meta-platforms, and Google Parent Alphabet. Apple's shares rose 1% Monday morning and are now up

36% over the past 12 months, well above the 21% gain averaged by those other four, end quote. From the If You Can't Beat Em, Join Em, File, Neiman Lab takes a look at how some journalists are taking freelance jobs with AI training data companies like Scale AI, which recruit them for tasks such as fact-checking and prompt drafting. Quote, for the past couple months, McKenna has been working close to full-time for Outlier, picking up projects on its gig platform at about $35 per hour. Data work has quickly become her primary source of income and a hustle she's recommended to other middil classmates. A lot of us are still looking for jobs. Three times I told someone what I do, and they're like, please send it to me, she said,

it's hard right now, and a lot of my colleagues are saying the same thing. McKenna is just one of many journalists who has been courted by Outlier to take on part-time remote data work over the past year. I spoke to local newswriters, photojournalists, and radio reporters across the U.S. who received similar recruitment messages from the company or heard about the platform through word of mouth among freelance journalists. Several of them told me they have taken on Outlier projects to supplement their income or replace their work in journalism entirely

because of dwindling staff jobs or freelance assignments drying up. Some are early career journalists like McKenna, but others are reporters with over a decade of experience. One thing they all had in common? Before last year, they'd never heard of Outlier or even knew that this type of work existed. Launched back in 2023, Outlier is a platform owned and managed by Scale AI, a San Francisco-based data annotation company, valued at $13.8 billion. It counts among its customers the world's largest AI companies, including OpenAI, Meta, and Microsoft. outlier and similar platforms like Crowded Gen and Remote Tasks use networks of remote human workers to improve the AI models of their clients. Workers are paid by the hour for tasks like labeling

training data, drafting test prompts, and grading the factual accuracy and grammar of outputs. Often their work is fed back into an AI model to improve its performance through a process called reinforcement learning with human feedback, or RLHF. This human feedback loop has been core to building models like OpenAIs GPT and MetaSlamma. Aside from direct recruitment messages, I also found dozens of recent public job postings that underscore this growing trend of hiring journalists for data work. These posts came from the AI industry's leading training data companies, including AAPN, data annotation, and scale AI itself. All of the openings list journalists as preferred candidates, often alongside editors, copy editors, and technical writers. Many job posts I found

are looking for language experts, including journalists who speak languages and dialects less represented in the training data of major AI companies. I found postings, I found postings. I found posts for fact-checkers internationally who speak Thai, Dutch, Hindi, and Swedish, as well as dialects like Spanish, Mexico, and French Canada. English-speaking journalists tended to qualify for more generalist job postings. These were often listed with titles like AI writing, evaluator, freelance writer, and fact-checker, end quote. Finally, have you heard the term Tiny Team? Let me hip you to it. Basically, some folks are coming to believe that AI is making startups possible with like only one or two people, and that's

it. Quoting the times. Gamma is among a growing cohort of startups, most of them working on AI products that are also using AI to maximize efficiency. They make money and are growing fast without the funding or employees they would have needed before. The biggest bragging rights for these startups are for making the most revenue with the fewest workers. Stories of tiny team success have now become a meme with techie's excitedly sharing lists that show how any sphere, a startup that makes the coding software cursor, hit $100 million in annual recurring revenue in less than two years with just 20 employees, and how 11 labs, an AI-voice startup, did the same with around 50 workers. With AI tools, some startups are now declaring that they

will stop hiring at a certain size. Runway Financial, a finance software company has said it plans to top out at 100 employees because each of its workers will do the work of one and a half people. Agency, a startup using AI for customer service, also plans to hire no more than 100 workers. It's about eliminating roles that are not necessary when you have smaller teams, said Elias Torres, agency's founder. The idea of AI-driven efficiency was bolstered last month by Deepseek, the Chinese AI startup that showed it could build AI tools for a small fraction of the typical cost. Its breakthrough built on open-source tools that are freely available online, set off an explosion of companies building new products using Deepseek's

inexpensive techniques. Deep Seek was a watershed moment, said Garabh, Garab Jane, an investor at the venture firm of four capital, which has backed Gamma. The cost of compute is going to go down very, very quickly, end quote. Mr. Jane compared new AI startups to the wave of companies that arose in the late 2000s after Amazon began offering cheap cloud computing services. That lowered the cost of starting a company, leading to a flurry of new startups that could be built more cheaply. Before this AI boom, startups generally burned $1 million to get to $1 million in revenue, Mr. Jane said. Now getting to $1 million in revenue costs one-fifth as much and could eventually dropped to one-tenth, according to an analysis of 200 startups conducted by a four.

This time we're automating humans, as opposed to just the data centers, Mr. Jane said. But if startups can become profitable without spending much, that could become a problem for venture capital investors who allocate tens of billions to invest in AI startups. Last year, AI companies raised $97 billion in funding, making up 46% of all venture investment in the United States, according to Pitchbook, which tracked startups. Venture capital only works if you get money into the winners, said Terrence, Roan, the investor with otherwise fund, who focuses on very young startups. He added, if the winner of the future needs a lot less money because they'll have a lot less people, how does that change VC?

For now, investors continue to fight to get into the hottest companies, many of which have no need for more money. Scribe and AI productivity startup grappled last year with far more interest from investors than the $25 million it wanted to raise. It was a negotiation of what is the smallest amount we could possibly take on, said Jennifer Smith, Scribe's chief executive. She said investors were shocked at the size of her staff, 100 people, when compared with its 3 million users and fast growth. Some investors are optimistic that AI-driven efficiency will spur entrepreneurs to create more companies leading to more opportunities to invest. They hope that once the startups reach a certain size, the firms will adopt the old model of big teams and big money. Some young companies,

including AnySphere, the one behind Cursor, are already doing that. AnySphere has raised $175 million in funding with plans to add staff and conduct research, according to the company. President Oscar Schultz, end quote. This podcast is a tiny team of one. Talk to you tomorrow.