Tech Brew Ride Home - Tue. 06/25 – AI Music Startups Get The Napster Treatment

Episode Date: June 25, 2024

History is rhyming today as AI music startups get the Napster treatment and Microsoft gets dinged for product bundling. Google wants you to build your own AI celebrity. Amazon wants to go at ChatGPT directly. And two hella-interesting and hella-big AI raises. Sponsors: Miro.com Links: Major record labels sue AI company behind ‘BBL Drizzy’ (The Verge) EU charges Microsoft with antitrust violations over Teams (FT) Apple Spurned Idea of iPhone AI Partnership With Meta Months Ago (Bloomberg) Uber Is Locking Out NYC Drivers Mid-Shift to Lower Minimum Pay (Bloomberg) Google Develops Challenger to Meta’s Chatbots and Character.AI (The Information) Amazon is secretly working on a ChatGPT killer (Business Insider) Etched is building an AI chip that only runs one type of model (TechCrunch) EvolutionaryScale lands $142 mln to advance AI in biology (Reuters) 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, June 25th, 2024. I'm Brian McCullough today. History is rhyming today as AI music startups get the Napster treatment and Microsoft gets dinged for product bundling. Google wants you to build your own AI celebrities. Amazon wants to go at chat GPT directly and two hella interesting and hella big AI raises. Here's what you miss today in the world of tech. The recording industry of America has sued AI music services, Suno, and you. UDio over alleged mass copyright infringement and claims they are trying to, quote, hide the full

scope of their infringement. Quoting the Verge. A group of record labels including the Big Three, Universal Music Group, Sony Music Entertainment, and Warner Records are suing two of the top names in generative AI music making, alleging the companies violated their copyright en masse. The two AI companies, Suno and UDio, used text prompts to churn out original songs. Both companies have enjoyed a level of success. Suno is available for use in Microsoft co-pilot through a partnership with the tech giant,

UDio was used to create BBL Drizzi, one of the more notable examples of AI music going viral. The case against Suna was filed in Boston federal court and the UDio case was filed in New York. The labels say artists across genres and eras had their work used without consent. The lawsuits were brought by the Recording Industry Association of America, the powerful group representing major players in the music industry, and a group of labels. The R-IAA is seeking damages of up to $150,000 per work, along with other fees. In its complaints, R-I-A-A included several examples of outputs generated using Suno and Udio that sound like songs owned by labels.

One song generated by Suno titled, Deep Down in Louisiana Close to New Orleans, by the way, that's how it was spelled, replicates the lyrics and style of Johnny Be Good by Chuck Berry. Another song called Prancing Queen generated using the prompt 70s pop contains lyrics to Dancing Queen by Aba and sounds remarkably like the band, end quote. Someone said online last night that they have thus far refrained from investing in music AI startups because if there's any industry that is organized and litigious when it comes to technological disruption to its industry, it is the music industry. They've got like 25 years experience in this.

Just ask Napster. Speaking of the year 1999, calling and wanting its headlines back, it has happened. In preliminary findings, the EU has charged Microsoft with antitrust violations over concerns that it gave teams an undue advantage by bundling it with office. Sound familiar to anyone quoting the Financial Times. The charges are the biggest that Brussels has filed against Microsoft, apart from merger control, since the group's showdown with the U.S. and EU over Windows, which began more than 20 years ago. That landmark case also centered on Microsoft's bundling or tying of its various software products.

In April, Microsoft issued concessions aimed at trying to avert regulatory action, including widening plans to unbundled teams from other software such as Office Beyond Europe. However, officials did not think the move went far enough to enable competition in the market. Reacting to the fresh charges, Brad Smith, Microsoft President said, having unbundled teams and taken initial interoperability steps, we appreciate the additional clarity provided today and will work to find solutions to address the Commission's remaining concerns. Sebastian Niles, President at Salesforce, which lodged the original complaint against Microsoft said the charges against Microsoft were a, quote, win for customer choice,

and an affirmation that Microsoft's practices with teams have harmed competition. The Commission. The EU's Executive Arms said fresh charges did not prejudge the outcome of the probe. Microsoft is seeking to settle the case to avoid formal charges that it is breaking the law and a potentially hefty fine of up to 10% of its annual global revenues, according to people familiar with the company's thinking, end quote. Mark German says, no, Apple is not looking to integrate Meta's Lama into the iPhone, though it's unclear if that is separate from integrating meta's various chatbots, as I told you about yesterday, quoting Bloomberg. The two companies aren't in discussions about using Meta's.

Lama Chatbot in an AI partnership and only held brief talks in March said the people who asked not to be identified because the situation is private. The dialogue about a partnership didn't reach any formal stage, and Apple has no active plans to integrate Lama. The preliminary talks occurred around the time Apple started hashing out deals to use OpenAIs, chat GPT, and Alphabet's Gemini in its products. Apple decided not to move forward with formal meta discussions in part because it doesn't see that company's privacy practices as stringent enough, according to the people. has spent years criticizing meta's technology and integrating Lama into the iPhone would have been a stark about face, end quote. And another follow-up story here, Uber has apparently begun

locking New York City drivers out of its app during low demand periods to fight that local rule that requires drivers to be paid for idle time between rides, quoting Bloomberg. At the heart of the move, say the two companies is a six-year-old pay rule in New York that, among other things, requires firms like Uber and Lyft to pay drivers for the idle time they rack up between rides. The lockouts, which began last month, are aimed at limiting how much non-passenger time drivers are able to log and be paid for. Drivers, meanwhile, say they need to work longer hours to earn the same amount as before. The lockouts occur unpredictably, making it difficult for drivers to plan work shifts and treat Uber as a full-time job. Sometimes these episodes can last

over an hour. An Uber spokesperson said that access to the platform is based on rider demand at any given time and place. If demand drops too far below supply, the company will temporarily shut drivers out. Nicholas Tejukti, who drives full-time for Uber, said that he has been getting shut out of the app four or five times a day. I used to work 10 hours and make $300 to $350, Tajuki said. Now I worked 10 hours and barely made $170. I was so disappointed. I'm paying for my gas and cannot make money. Wesley Dorsenaville, another full-time Uber driver, similarly said he used to take home between $300 and $400 per shift, but was lately seeing between $170 and a little over $200, end quote.

But back to AI. Sources say Google has been developing a product to create customizable chatbots, which could be modeled on celebrities and plans to launch it as soon as later this year, quoting the information. The search engine giant has been developing a product for creating and conversing with customizable chatbots, which could be modeled on celebrities or made by users to people with direct knowledge of the project said, the bots would be similar to the online personalities from meta platforms and startup character AI based on celebrities like football star Tom Brady

and TV character Tony Soprano. Ryan Germick, a longtime designer at Google, known for his work on the Google Doodles that appear on the company's search homepage, is leading the project as part of the Google Labs team, which works on experimental AI products, the people said. Around 10 Google employees have been working on it. The chatbot will likely appear on the Google Labs website if the company makes it accessible to the public. As currently conceived, Google's Gemini AI model would power the chatbots. Users would be able to create their own chatbots by describing its personality and

appearance. That feature is similar to character AI, which also allows users to give the chatbots elaborate backstories and to specify what their voices sound like in audio conversations. As part of the effort, Google has discussed striking partnerships with influencers to create chatbots based on them, one of the people said. Google plans to launch the feature as a standalone product, but has also discussed eventually integrating it into YouTube, this person said. YouTube leaders have said they want to harness generative AI advances to help creators

who frequently say they would like more ways to engage with fans, end quote. Sources tell Business Insider that Amazon is working on an AI chatbot codenamed Medis designed to compete directly with ChatGPT. One source says it's tentatively slated for a September launch, quote, the secret internal project is codenamed Medis, likely in reference to the Greek goddess of wisdom. The new service is designed to be accessed from a web browser similar to how other AI assistants work, according to people familiar with the project and an internal document obtained by BI. Medus is powered by an internal Amazon AI model called Olympus, another name inspired by Greek mythology.

This is a more powerful version of the company's publicly available Titan model, the people familiar said. At the most basic level, Medis gives text and image-based answer. in a smart conversational manner, according to the internal document. It also is able to share links to the source of its responses, suggest follow-up queries, and generate images. Amazon wants METIS to use an artificial intelligence technique called Retrieval Augmented Generation, or Rag, the people said, this means METIS will be able to retrieve information from beyond the original data used to train its underlying Olympus model. The goal is to generate more up-to-date responses. For example, METIS should be able to share the latest stock prices, while some other chatbots

that are not rag can't do so, the people familiar with the situation said. Medis is also expected to work as an AI agent, one of the people said. AI agents are capable of automating and performing complex tasks based on existing data, like making a vacation itinerary. Medus's use cases could include turning on your lights and booking a flight for you, one of the people told BI, end quote. Also on the Amazon tip, just noting this real quickly so you can mark your calendars, Amazon plans to hold Prime Day 2024 on July 16th and July 17th.

in 24 countries, including the U.S., the UK, and Japan. India will receive a separate event later in the year. Finally today, a couple of interesting raises in the AI space. Etched, which is building Sohu, an inferencing ship that only runs transformer AI models, has raised a $120 million Series A, bringing its total funding to $125.36 million to date, quoting TechCrunch. Only two years old, etched was founded by a pair of Harvard dropouts, Gavin Uberti, X-OctoML, and X-XNOR AI, and Chris Zhu, who along with Robert Wachin and former Cypress

Semiconductor CTO Mark Ross, sought to create a chip that could do one thing. Run AI models. That's not unusual. Plenty of startups and tech giants have or are developing chips that exclusively run AI models, also known as inferencing chips. Meta has MTIA, Amazon has Graviton and Inferentia, and so on. But etched chips are unique in that they only run a single single. type of model. Transformers. The Transformer proposed by a team of Google researchers back in 2017 has

become the dominant generative AI model architecture by far. Transformers underpin OpenAIs video generating models. They're at the heart of text generating models like Anthropics Claude and Google's Gemini. And they power art generators, such as the newest version of stable diffusion. In 2022, we made a bet that Transformers would take over the world. Huberti etch's CEO told TechCrunch in an interview. We've hit a point in the evolution of AI where specialized chips that can perform better than general-purpose GPUs are inevitable, and the technical decision-makers of the world know this. Etch's chip, called Sohu, is an ASIC application-specific integrated circuit, a chip tailored for a particular application, in this case running transformers. Manufactured using TSM's four-nanometer

process, So-HU can deliver dramatically better inferencing performance than GPUs and other general-purpose AI chips, while drawing less energy, claims UberTI. Sohu is an order of magnitude faster and cheaper than even NVIDIA's next generation of Blackwell GB200 GPUs when running text, image, and video transformers, Uberti said. One Sohu server replaces 160 H-100 GPUs, So-H-100 GPUs. So-who will be a more affordable, efficient, environmentally friendly option for business leaders that need specialized chips, end quote. How does So-who achieve this?

In a few ways, but most obvious and intuitive, is a streamlined inferencing hardware and software pipeline. Because so who doesn't run non-transformer models, the etched team was able to do away with hardware components not relevant to transformers while trimming the software overhead traditionally used to deploy and run non-transformers. While etched lacks a direct competitor at present, AI-chip startup perceived recently previewed a processor with hardware acceleration for transformers as well. GROC has also invested heavily in transformer-specific optimizations for its ASIC. competition aside, what if Transformers one day fall out of favor? Uberti says that in that case, Etch will do the obvious, design a new chip.

Fair enough, but that's a pretty drastic fallback, considering how long it's taken to bring Sohu to fruition, end quote. And finally, finally, a startup called Evolutionary Scale has released AI models called ESM3 to create novel proteins. To do so, it has raised $142 million in seed funding led by, who else? our friend Nat Friedman, Daniel Gross, and Lux Capital, quoting Reuters. Amazon Web Services and the venture capital arm of Invydia also participated in the fundraising, calling it a chat GPT moment for biology, Lux Capital co-founder and managing partner Josh Wolfe, told Reuters the company has developed the first large language model for creating novel proteins and other biological systems. Evolutionary scale envisions its AI being utilized for a wide range of applications

from accelerating drug discovery to engineering microbes capable of breaking down plastic in the environment the company's chief scientist Alex Reeves told Reuters. Using AI to engineer new biological systems is currently an area of intense interest. At the same time, experts have also raised alarm bells about generative AI's potential to help create bioweapons through aiding the development of harmful pathogens or toxins. Evolutionary scales funding will be used to train its next generation of AI models, as well as build out a team to partner with the biotech industry. Reeves said. The company is releasing models named ESM3. The smaller ESM3 model is being open-sourced

for non-commercial research, and AWS and NVIDIA will make the models available commercially, including its largest ESM3 model the company said. The company said it leveraged ESM3 to engineer a novel fluorescent protein that diverged from the evolutionary path of naturally occurring fluorescent proteins, which would have taken nature about 500 million years to evolve, end quote. Here's Lux's Josh Wolf in a tweet storm from this morning, quote. Biology is the most advanced technology that has ever been created, far beyond anything that people have engineered. The ribosome is programmable. It takes the code of proteins in the form of RNA and builds them up from scratch, fabrication at the atomic scale. Every cell and every

organism on Earth has thousands to millions of these molecular factories. But even the most sophisticated computational tools created to date barely scratched the surface. biology is written in a language we don't yet understand. If we could learn to read and write in the code of life, it would make biology programmable. trial and error would be replaced by logic and painstaking experiments by simulation. ESM3 is the first generative model for biology that simultaneously reasons over the sequence, structure, and function of proteins. ESM3 is trained across the natural diversity of the earth, billions of proteins from the Amazon rainforest to the depths of the oceans, extreme environments like hydrothermal vents and the microbes in a handful of

soil. We believe that ESM3 is the most parameters plus data plus compute ever applied to training a biological model, trained with over one times 1,024 flops and 90 billion parameters. Trained on one of the highest throughput GPU clusters in the world today, ESM3 is a frontier generative model for biology, end quote. Nothing more for you today, talking about. to you tomorrow.