Tech Brew Ride Home - Tue. 04/18 – Is The AI Era Kicking Off A New Generation Of Silicon?

Episode Date: April 18, 2023

Is the AI revolution going to kick of a silicon revolution with a whole new generation of chips? Coinbase would consider leaving the US if it can’t get clarity from regulators. Is the Apple savings account the savings account for you? A look at Twitter competitor Bluesky. And what happens if you take GPT-4 and feed it back on itself? Sponsors: Notion.com/ride Grammarly.com/techmeme Links: Microsoft Readies AI Chip as Machine Learning Costs Surge (The Information) Coinbase Could Move Away From U.S. if No Regulatory Clarity: CEO Brian Armstrong (CoinDesk) Major Photography Prize Winner Reveals Image Is AI-Generated, Rejects Award (Motherboard) Apple finally launches high-yield Apple Savings account (Apple Insider) Bluesky’s CEO wants to build a Musk-proof, decentralized version of Twitter (The Verge) Hype grows over “autonomous” AI agents that loop GPT-4 outputs (ArsTechnica) Learn more about your ad choices. Visit megaphone.fm/adchoices

Welcome to the TechMame right home for Tuesday, April 18th, 2023. I'm Brian McCullough today. Is the AI revolution about to kick off a silicon revolution with a whole new generation of chips? Coinbase would consider leaving the U.S. if it can't get clarity from regulators. Is the Apple's savings account, the savings account for you? A look at Twitter competitor, Blue Sky, and what happens if you take GPT4 and feed it back on itself? Here's what you missed today in the world of tech. So, interesting thing about this AI revolution we've already seen. it upsetting and disrupting almost every corner of the tech universe. But guess what? It looks like

it's going to disrupt the chip sector, too. What if this AI era leads to a whole new category of chips or sets off a whole new chip horse race with a whole new generation of silicon? Sources are telling the information that Microsoft is developing an in-house AI chip, internally codenamed Athena, and has been doing so since as early as 2019. They are apparently far enough along already as to be testing it with some Microsoft and OpenAI employees. Quote, after placing an early bet on OpenAI, the creator of ChatGBTGPT, Microsoft has another secret weapon in its arsenal, its own artificial intelligence chip for powering the large language models responsible for understanding and generating human-like language.

Microsoft is hoping the chip will perform better than what it currently buys from other vendors, saving it time and money on its costly AI efforts. Other prominent tech companies, including Amazon, on Google and Facebook also make their own in-house chips for AI. The chips which are designed for training software such as large language models, along with supporting inference, when the models use the intelligence they acquire in training to respond to new data, could also relieve a shortage of the specialized computers that can handle the processing needed for AI software. That shortage, reflecting the fact that primarily just one company, Nvidia, makes the chips, is felt all across tech. It has forced Microsoft to ration its computers for some

some internal teams the information has reported. According to Tracy Wu, a senior cloud analyst at Forrester Research, the AI boom is putting increasing pressure on cloud providers to work on their own chips. You can buy from Nvidia, but when you're looking at these huge behemists like Google and Amazon, they have the capital to build out and design their own chips, she said. Microsoft could make Athena widely available for use within the company itself and OpenAI as early as next year, the two people said. However, Microsoft is still debating whether it it will make the chips available to customers of its Azure Cloud Computing service, according to one of the people.

That's because most customers don't need to train their own LLMs and might not require the chip's horsepower, they said. If Microsoft does decide to make Athena available to its cloud computing customers, it will also have to build software for it that is more compelling than Nvidia's current offerings, which Nvidia has been improving on for 15 years and which developers now widely use. Dylan Patel, chief analyst at research firm Semi-Analysis, pegged the the cost of operating chat GPT at around $700,000 a day, or 36 cents per query. Most of this cost is based around the expensive servers they require, he said.

Athena, if competitive, could reduce the cost per chip by a third when compared with NVIDIA's offerings. Microsoft wants to use large language models across all their applications, including Bing, Office 365, and GitHub, he added, to deploy that at scale using off-the-shelf hardware would cost tens of billions of dollars a year, end quote. Coinbase CEO Brian Armstrong says, anything is on the table, including relocating the company from the U.S. if U.S. regulatory clarity for the industry does not emerge, quoting CoinDesk. I think the U.S. has the potential to be an important market for crypto, but right now we are not seeing that regulatory clarity that we need, he said.

I think in a number of years, if we don't see that regulatory clarity emerge in the U.S., we might have to consider investing more elsewhere in the world, end quote. Armstrong's comments came weeks after rival exchange, Bittrex, said it planned to exit the U.S. by the end of April, citing, quote, the current U.S. regulatory and economic environment, end quote. Bitrex received a Wells notice, a statement that the U.S. Securities and Exchange Commission's enforcement division found evidence of legal violations in March, General Counsel David Maria told the Wall Street Journal. The SEC filed a lawsuit against the exchange on Monday. Armstrong compared the U.K. situation where there is only one regulatory. later, the Financial Conduct Authority, or FCA, responsible for both commodities and securities,

with the U.S. where there are separate bodies, the Commodity Futures Trading Commission, and the SEC. You don't have this unfortunate thing happening where the CFTC and the SEC are having a turf battle, he said. We actually have contradictory statements from the heads of the CFTC and the SEC coming out almost every few weeks. How's a business going to operate in that environment? We just want a clear rulebook, end quote. Coinbase received a Wells notice from the SEC in March. Armstrong said Coinbase had met with the SEC 30 times without getting feedback regarding the nature of its business before receiving the notice, end quote. Apple finally launched its Apple Card Savings account, announced all the way back in October of 2022,

letting users deposit daily cash rewards into a 4.15% high yield account with no fees. Quoting Apple Insider, savings helps our users get even more value out of their favorite Apple Card benefit, daily cash, while providing them with an easy way to save money every day, said Jennifer Bailey, Apple's vice president of Apple Pay and Apple Wallet in a statement. Our goal is to build tools that help users lead healthier financial lives and building savings into Apple Card and wallet enables them to spend, send, and save daily cash directly and seamlessly, all from one place, end quote. Apple says that the 4.15% APY rate is, quote, more than 10 times the national average. The company also says that Apple Savings has no fees, no minimum deposits, and no

minimum balance requirements. In a recent survey, Fortune Magazine reported that the U.S. national average is 0.37% APY, annual percentage yield. However, it also recommended 10 savings accounts, every one of them, which offers more than Apple. Those range from Lending Club with 4.25% APY to UFB Direct with 4.81% APY. Fortune also notes that APY is not the sole factor in determining the right savings account to invest in. There are also issues to do with minimum deposits, balances, and customer support, Apple savings is likely to prove particularly strong on customer support because it launches with a savings dashboard in the wallet app. Apple says this means users, quote, can conveniently track their account balance and interest earned over time, end quote.

The launch of Apple savings comes after an announcement in October 2022 and then mostly silence since. A line about Apple savings was added to the Apple card terms and conditions in early March 23. Then most recently, code showing the backend service was live was spotted. It's not clear why the service launched so long after the announcement, but it's possible that it is because the finances are being serviced by Goldman Sachs, Apple's partner in the Apple card. Despite the apparent success of Apple Card, Goldman Sachs posted a $1.2 billion loss, which was attributed in part to its consumer offerings, such as that with Apple. Consequently, Goldman Sachs recently pulled out of doing any other consumer credit card. alongside that, the Federal Reserve opened an investigation into whether Goldman Sachs's Consumer Credit Division has sufficient consumer protections. Now that Apple Savings is here,

it's intended that users will set it up once and then not have to deal with it until they choose to. Apple Card users continue to get rewards as a percentage of their purchasing, and they can continue to have that go into their existing daily cash account. However, if they choose to, those payments can instead go to the new high-yield savings account, end quote. But also, you can just, you know, transfer cash in there directly. I did, in fact, shift my Apple Rewards Daily Cash to go over to this new savings account, but I'm also shifting some money over there, too, from other savings accounts, because, you know, that APY is more than I'm currently getting at Ally, so why not? A winner at the Sony World Photography Awards has given the award

back, saying his submission was AI generated and was meant to test if such competitions were ready for AI. Quoting motherboard. In March this year, the World Photography Organization announced the winners of the Sony World Photography Awards, 2023. Among many stunning images that took home various prizes was pseudamnesia, the electricia by German artist Boris Eggleston, which won the creative category. A Sony press release described the image as, quote, a haunting black and white portrait of two women from different generations reminiscent of the visual language of 1940s family portraits, end quote. However, the image was generated by AI and not a real photo, something that is obvious when you zoom in and look at the hands in the image, for example. On his personal website,

Eggleston revealed the truth and refused the award, quote, I applied as a cheeky monkey to find out if the competitions are prepared for AI images to enter. They are not, he wrote. We, the photo world, need an open discussion, a discussion about what we want to consider photography and what not. Is the umbrella of photography large enough to invite AI images to enter, or would this be a mistake? With my refusal of the award, I hope to speed up this debate, end quote. Eglson wrote that he has been a photographer for 30 years and recently turned to AI as part of his artistic practice. The winning image is, quote, a complex interplay of prompt engineering in painting and outpainting that draws on my wealth of photographic knowledge, he wrote.

A world photography organization spokesperson told motherboard that, while the judges were aware that AI was used in creating the image, they were under the impression that it was a co-creation with AI and that Eggelson, quote, deliberately misled them, end quote. Twitter competitor watch, with the usual caveat that the right home fund is an investor in Twitter competitor T2, the verge takes a look at Blue Sky, that attempt led by Jack Dorsey to create an open source protocol known as the AT protocol. In an interview, Blue Sky CEO, Jay Graber, revealed that Blue Sky has 20,000 users so far. Quote,

Before Elon Musk reluctantly bought Twitter, former CEO Jack Dorsey funded and spun off Blue Sky, a public benefit company tasked with building an open-source decentralized social media protocol that he wanted Twitter itself to eventually operate on. Blue Sky remains invite only to access, but is now starting to let in more users. After setting up my account recently, I wanted to talk with CEO Jay Grabber. She agreed to answer some of my questions this week for what I believe is her first interview since she joined to lead the project in August 2021. Right now,

Blue Sky's interface is a shameless clone of Twitter. A key difference is that it defaults to a chronological feed of who you follow and lets you choose to toggle between a what's hot algorithmic feed. The service currently lacks the basic tools it needs to live up to its decentralized mission, including the ability to export account data. Big parts of what the underlying AT protocol promises to deliver such as a marketplace of feed algorithms to choose from also don't exist yet. The company's approach to content moderation was just outlined its early days. Still, I am fascinated by what Blue Sky will become. In the near term, it represents direct competition to Mastodon and Noster, another decentralized social networking protocol that is also funded by Dorsey.

In the long term, the ethos behind Blue Sky could signal where the rest of the social media world is headed, end quote. Click through to read the interview. I actually haven't had a chance to sign up to Blue Sky myself, though. someone just sent me an invite, so I'll do that soon and report back. And finally, Ars Technica takes a look at autonomous AI agents, like Auto-GPT and Baby AGI, which perform multi-step tasks by iteratively taking output from GPT4 and then feeding it back into itself. Quote, since the launch of OpenAI GPT4 API last month to beta testers, a loose group of developers has been experimenting with making agent-like or agentic implementations of the AI model that attempt to carry out multi-step tasks with as little human intervention as possible.

These homebrew scripts can loop, iterate, and spin off new instances of an AI model as needed. Two experimental open-source projects in particular have captured much attention on social media, especially among those who hype AI projects relentlessly. One, auto-GPT, created by Torin Bruce Richards, and baby AGI created by Yohei Nakajima. What do they do? Well, right now, not very much. They need a lot of human input and handholding along the way, so they're not yet as autonomous as promised,

but they represent early attempts toward more complex chaining AI models that could potentially be more capable than a single AI model working alone. Basically, auto-GPT takes output from GPT4 and feeds it back into itself with an improvised external memory so that it can further iterate on a task, correct mistakes or suggest improvements. Ideally, such a script could serve as an AI assistant that could perform any digital task by itself. To test these claims, we ran AutoGPT, a Python script locally on a Windows machine. When you start it, it asks for a name for your AI agent, a description of its role, and a list of five goals it attempts to fulfill. While setting it up, you need to provide an open AI API

key and a Google search API key. When running, AutoGPT asks for permission to perform every step it generates by default, although it also includes a fairly automatic mode if you're feeling adventurous. If tasked to do something like purchase a vintage pair of Air Jordans, AutoGPT will develop a multi-step plan and attempt to execute it. For example, it might search for shoe sellers, then look for a specific pair that meets your criteria. But that's when it stops because it actually can't buy anything at the moment. If hooked into an appropriate purchasing API, that could be possible. If you want to get a taste of what AutoGPT does yourself, someone created a web-based version called Agent GPT that functions in a similar way.

Richards has been very open about his goal with Auto-GPT to develop a form of AGI or artificial general intelligence. In AI, general intelligence typically refers to the still hypothetical ability of an AI system to perform a wide range of tasks and solve problems that are not specifically programmed or trained for. Like a reasonably intelligent human, a system with general intelligence should be able to adapt to new situations and learn from experience, rather than just following a set of predefined rules or patterns. This is in contrast to systems with narrow or specialized intelligence, sometimes called narrow AI, which are designed to perform specific tasks or operate within a limited range of contexts. Meanwhile, baby AGI, which gets its

name from an aspirational goal of working toward artificial general intelligence, work, in a similar way to Auto-GPT, but with a different task-oriented flavor. You can try a version of it on the web at a site not so modestly titled God Mode. Nakajima, the creator of Baby AGI, tells us that he was inspired to create his script after witnessing the Hustle GPT movement in March, which sought to use GPT4 to build businesses automatically as a type of AI co-founder, so to speak. It made me curious if I could build a fully AI founder, Nakajima says. why Auto-GPT and baby AGI fall short of AGI is due to the limitations of GPT-4 itself. While impressive as a transformer and analyzer of text, GPT4 still feels restricted to a narrow range

of interpretive intelligence, despite some claims that Microsoft has seen sparks of AGI-like behaviors in the model. In fact, the limited usefulness of tools like Auto-GPT at the moment may serve as the most potent evidence yet of the current limitations of large language models. Still, that does not mean those. limitations will not eventually be overcome. When asked for a useful application of Baby AGI, Nakajima couldn't come up with a substantive example aside from a do-anything machine, a project built by Garrett Scott that aspires to create a self-executing to-do list, which is currently in development. To be fair, the Baby AGI project is only about a week old. It's more of an introduction

to a framework slash approach, and what's most exciting are what people are building on top of this idea, he says, end quote. Hey folks, ride home fund has a portfolio company that is looking to hire its first ever VP of engineering. The company is denim, and they haven't done a portfolio profile episode yet, but you can check them out at denim. That's d-enim.us. If you think you'd be a good candidate for this role and are interested, email me at brian at ridehomefund.com, and I'll forward you along to the company. Talk to you tomorrow.