Tech Brew Ride Home - Wed. 04/17 – Copilots Or The Biggest Model Wins? The Competing AI Visions.

Episode Date: April 17, 2024

T-Mobile and Verizon employees are being offered bribes for SIM swaps. How much coding copilots have taken over. How much would you value Mistral or Cohere in the LLM race? How popular has Airchat gotten? How big has Amazon Prime gotten? And one of the original modern robots is being retired. Links: T-Mobile, Verizon workers get texts offering $300 for SIM swaps (BleepingComputer) Microsoft’s AI Copilot Is Starting to Automate the Coding Industry (Bloomberg) Mistral, an OpenAI Rival in Europe, in Talks to Raise Capital at a $5 Billion Valuation (The Information) Invitation-Only Audio Social Network Is the Hot New App in Tech Circles (Bloomberg) Musk’s Starlink Cracks Down on Growing Black Market (WSJ) Amazon Prime Memberships in US Gain 8% to New High After Lull (Bloomberg) Atlas shrugged: Boston Dynamics retires its hydraulic humanoid robot (TechCrunch) 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 Wednesday, April 17th, 2024. I'm Brian McCullough today. T-Mobile and Verizon employees are being offered bribes for sim swaps.

How much coding co-pilots have taken over? How much would you value mistral or cohere in the LLM race? How popular has AirChat gotten? How big has Amazon Prime gotten? And one of the original modern robots is being retired. Here's what you miss today in the world of tech. This could be just random hacker activity, but I keep feeling like we're seeing strange strategic hacks all of the sudden, suggesting

someone nation-stady is maybe laying the groundwork for something big. Current and former T-Mobile and Verizon employees say they are receiving texts offering them up to $300 to perform a SIM swap. T-Mobile says it is investigating, quoting bleeping computer. According to many reports, this is part of a campaign targeted at current and former mobile carrier workers who could have access to the systems required to perform a sim swap. Myself and a group of former employees all got the same message all from different numbers and with different verbiage, one of the targeted people said.

The attacker's messages claim the source of the contact information is an employee directory as first spotted by the mobile report. I got your number from the T-Mobile employee directory. I'm looking to pay someone up to $300 per sim swap done. If you're interested, reply and we can talk, the criminals say in the messages. Other texts seen by bleeping computer requests employees contact the threat actor on telegram if they are interested in the offer. While it was initially believed that these texts were only targeting T-Mobile employees, Verizon employees also stated that they were receiving variations of the same text.

T-Mobile told Bleeping Computer that they are investigating the text messages, but stressed they did not suffer another breach. We did not have a systems breach. We continue to investigate these messages that are being sent to solicit illegal activity. we understand other wireless providers have reported similar messages, T-Mobile told, bleeping computer. In SimSwap attacks, criminals trick a victim's wireless carrier into rerouting their service to a device controlled by the fraudster. Successful attacks can lead to unauthorized access to personal accounts and sensitive information, identity theft, financial losses, and significant emotional

distress for the victims, end quote. The thing is, think of the threat surface something like this reveals. How many people are out there working mid to low-level jobs that nonetheless have access and keys to incredibly sensitive and valuable pressure points? They could easily be susceptible to bribery at the very least. I'm thinking of everyone from Apple employees to Verizon employees to utility power plant workers. The days when in order to hobble a nation state, you had to only target government employees or military personnel, those days are long over. We know that co-pilots like GitHub co-pilot have become popular, but how popular? According to Stack Overflow, 54.8% of developers used co-pilot in 2023, quoting Bloomberg. Co-Pilot is gradually revolutionizing

the working lives of software engineers, the first professional cohort to use generative AI en masse. Microsoft says Co-Pilot has attracted 1.3 million customers so far, including 50,000 businesses ranging from small startups to corporations like Goldman Sachs, Ford, and Earned. Ernst and Young. Engineers say co-pilot saves them hundreds of hours a month by handling tedious and repetitive tasks, affording them time to focus on naughtier challenges. Acquired by Microsoft in 2018 for $7.5 billion, GitHub dominates its market, and as betting, Copilot has the AI horsepower to fight off rival services, including Tab 9, Amazon's Code Whisperer, and Google-backed Replet Ghostwriter. GitHub's AI assistant is also a kind of beta test for a host of other co-pilots

that Microsoft is baking into office, Windows, Bing, and other business lines. As is true with AI generally, GitHub copilot has limitations. Developers say it sometimes pulls up outdated code, provides unhelpful answers to questions, and generate suggestions that are buggy or could infringe copyright. Because the tool is trained on public and open repositories of code, engineers run the risk of replicating security issues or injecting new ones into their work, particularly if they blindly accept co-pilot's recommendations. GitHub emphasizes that the tool is an assistant, not a substitute for human programmers, and has put the onus on customers to use it wisely. Robust guidelines are required to prevent lazy programmers from simply accepting what co-pilot suggests,

said GitHub chief executive officer Thomas Domk. He expressed confidence that engineers would keep one another honest. Some companies are starting to deploy co-pilot to create code for critical systems. Brewer-Karlsberg uses it to write code for an existing tool that helps the Salesforce plan, prepare for, and document sales calls. Mindful of co-pilot's limitations, the beer maker uses its own quality assurance process to check that the code it has created works as intended, according to Chief Information Officer Sarah Haywood. Eventually, she said, companies will be able to outsource that task as well. As time goes on, people will build more trust in AI, she said, I don't think we should be having to double-check everything that AI does, otherwise we're not really adding any value, end quote.

Few tasks are more tedious than debugging software, a process that can consume as much as 50% of engineers' time. Figma, which helps developers design app or website interfaces, says co-pilot can create defect testing programs in minutes rather than hours. That is the real value of AI, said Abyshech Mothar, the company's vice president of engineering. It doesn't replace our work, but frees up our time to develop creative solutions, end quote. There is an investing in AI thesis out there right now that suggests that this will be what the immediate AI revolution looks like. Co-pilots for everything tedious and wrote in every vertical you can imagine. Law, medicine, education, you name it. Now the converse to that thesis. Sources say mistral is talking to investors

to raise several hundred million dollars at a $5 billion valuation, quoting the information. The company, which has just recently begun to generate revenue in December raised $415 million at a valuation of $2 billion. The possible back-to-back financing reflects in the company. The company, which has just recently reflects investors continued appetite for certain AI startups and the large amount of capital, such startups need to compete with Google, Open AI, and other leaders in the field. It's unclear which investors Mistral has spoken to about a new funding round. Its existing investors include Andresen Horowitz, lightspeed venture partners, and Microsoft, which has also backed Open AI, but with far more capital.

A $5 billion valuation would stand out even among AI startups that have fetched steep prices on modest revenue, cohere, which also develops LIA, LLMs has been in talks to raise money at a $5 billion valuation despite generating revenue at an annualized pace of about $22 million. Mistral charges customers that use an application programming interface to access its large language models. The models, most of which are also available for free, have surged in popularity among application developers. Mistral is banking on its ability to convince AI customers in Europe to work with a locally based firm rather than with AI providers based in other parts of the world, end quote. That European angle is an interesting one,

but the reason I said this is the example of the inverse or converse of that previous AI investing thesis is that some folks out there are asking, what is the business model of a mistral? They're open sourcing their models. They require hundreds of millions of dollars just to train new models to keep up with the state of the art. But once they've done that, can they get enough paying customers to pony up for API? calls to cover the costs of training or even to justify that valuation. And there are, as mentioned, several other competitors out there doing largely the same thing. Yes, Open AI seems to be generating, or at least on the path to generating hundreds of millions of dollars a year in revenue. But note that

coherence of a mere $22 million. To bet big on the coheres and mistrills of the world, you need to believe the AI market will basically be so boundless and endless that it takes over all of compute. Oh, and it needs to do that without running. into any walls in the market like running out of quality data to train on or running out of compute. I was listening to an Oddlots episode yesterday where an analyst who focuses on electricity generation noted that the utility industry in North America estimates it needs to 2x the amount of electricity generated in this country just to keep up with AI. The lead time to spin up new power plants can be like a decade. What happens if we run out of the ability to power AI compute?

So you have the thesis of the AI Revolution will be co-pilots all the way down, or you have the thesis that the largest models always win and can do basically anything, and thus their total addressable market is basically infinite. Those are the two dominant theories VCs are operating under at the moment in AI. A bunch of follow-up stories real quickly, follow-ups to things we've discussed recently. According to Censor Tower, Naval Ravacant and Brian Norgard's audio phone, invitation-only social app air chat has been downloaded more than 30,000 times just since its relaunch earlier this month. Voting Bloomberg.

Ravikant said the point of the app isn't to catch up on what folks have been saying while you weren't staring at your phone. It's, quote, rather to see what's going on around you right now, he said, like dipping into conversations at a house party. Ravikant said that during an interview on air chat, saying he preferred to answer questions in public on the app. I want a house party in my pocket, he said. I want to be able to pull out this phone and talk to someone interesting and delightful and witty any time I want. AirChat has been downloaded more than 45,000 times worldwide since its initial launch in mid-20203, according to research firm Censor Tower. More than 30,000 of those downloads happened after it relaunched this month,

kicking off the current cycle of enthusiasm. Almost half of the downloads are from users based in the U.S. Over the last few days, air chat users have discussed topics ranging from the war in the Middle East to fitness and Brian Johnson's anti-aging quest. But the most prominent topic of discussion is air chat itself. Will it become the new dating app, opening up users to people they might have on first blush rejected if using Hinge or Tinder because they liked the person's voice and had a thoughtful conversation? Is AirChat going to be the next big thing?

End quote. Yes, for those of you asking, I did get an invite. Thank you, Swix. Turns out Chris Messina had also sent me one last week and I missed the message. I'm on there now. Naval and Brian, again, if you want to test putting podcast episodes up on there, let me know. And according to emails seen by the journal, SpaceX has begun cracking down on users who are connecting to its Starlink internet service from unauthorized countries, including Sudan, South Africa, and Zimbabwe. Quote, customers in those countries have received email

notifications from the company in recent days, warning that their access to the service would be terminated by the end of the month. The emails viewed by the journal noted that, using Starlink in areas where it hasn't been approved by local regulators was against the company's terms of service. The availability of our mobile service plans is contingent upon various factors, including regulatory approvals. The email said, referring to Starlink roaming products that allow subscribers to use its internet services in different countries. The notifications were sent just days after the journal published an investigation into the growing black market that has allowed users, including Russian military units fighting in Ukraine and a brutal militia in Sudan, to bypass local regulatory

restrictions on Starlink. Since launching the first operational Starlink satellites in 2019, SpaceX has been rushing to get regulators around the world to approve the service. By late March, 72 countries had authorized Starlink, according to the company. Jurisdictions that haven't approved the service include India and much of Africa. The service also isn't available in Russia and China, and SpaceX isn't seeking permission to offer Starlink there, according to the official Starlink availability map. But thousands of users across the globe have found ways around local restrictions, usually by purchasing Starlink kits in countries where the service is authorized and then signing up for one of the company's roaming packages. Middleman companies that offer to activate Starlink and send the kits to users in places where they aren't officially available have sprung up in countries like the United Arab Emirates and Mozambique, end quote.

And this is fascinating. We're all Amazon Prime members now. Well, mostly. According to CIRP, Amazon Prime hit a new high of a hundred, 180 million U.S. shoppers in March, up 8% year over year despite greater online competition. 75% of U.S. shoppers now have a prime membership, quoting Bloomberg. The increase shows Amazon's promise of fast delivery on a wide assortment of goods continues to have strong appeal with shoppers grappling with inflation, despite greater online competition from Walmart and China-linked companies such as TikTok shop, Shian, and Temu. The firm measures the number of people who have access to prime, which is great, than the actual number of subscriptions sold by Amazon. A prime subscription is generally shared by

members of a household, all of whom are counted by consumer intelligence research partners. Amazon in 2021 announced it had 200 million prime members globally the last time it released a figure on the number of subscriptions. The Seattle-based company discloses subscription services revenue each quarter, which is mostly from prime. Analysts project that Amazon will report a 12% gain to $10.8 billion in subscription service revenue in the period and in March 3rd. 31st, according to data compiled by Bloomberg, end quote. Finally, today, despite the recent surge in interest in robotics, one of the OGs is hanging up its hat. Boston Dynamics is retiring Atlas, that robot that has been basically the robot people post videos of when they want to post viral

videos of modern robots. Quoting TechCrunch, Boston Dynamics was, of course, well ahead of the current humanoid robotics curve. Last July marked the 10th anniversary of the bipedal robots debut. The company teamed with DARPA for Atlas's early development, leading the robot to be heavily incorporated into challenges of the era. Given the tremendous interest around companies like agility, figure, 1X, and Apptronic, it stands to reason that, at the very least, the Waltham, Massachusetts-based company has seriously explored the commercial humanoid category. Venturing an educated guess, I would say that today's Atlas' retirement is less about endings than new beginnings. Most likely, the hulking humanoid is gracefully stepping out of the way of whatever is coming. next for the firm. At the time of its debut, Atlas was one of the most advanced humanoid robots ever built DARPA notes, but it was essentially a physical shell for the software brains and nerves

that the teams developed. Atlas has taken many strides in the intervening decade, of course, as the bipedal robot continues to factor into Boston Dynamics' research and promotional material. Today, however, marks the end of the road for that robot. While many of the system's advances in locomotion still impress, certain aspects like its hydraulics are antiquated by contemporary robotic standards. Meantime, in lieu of a gold watch, Boston Dynamics is offering up a video featuring some of Atlas's greatest hits and most spectacular falls. It's one final blooper reel that demonstrates just how much work goes into those perfectly choreographed videos, end quote. Will people one day share videos of Atlas in the same way that people right now share videos of folks using giant

cell phones in the 1980s to be like, can you believe this was once considered cutting edge technology. Talk to you tomorrow.