- Synthetic
- Posts
- New Etched chip challenges Nvidia; The RIAA puts AI music on trial
New Etched chip challenges Nvidia; The RIAA puts AI music on trial
Plus, scientists create smiling robot face from living human skin
Steve Brown
June 27, 2024
In partnership with
Subscribe to Synthetic
This Week in AI
On Monday, the Record Industry Association of America (RIAA) filed lawsuits against AI music startups Suno and Udio. The lawsuit claims that the companies built their products using unauthorized scrapes of copyrighted works without the permission of artists and studios. The entire industry is eagerly watching the outcome. Will these companies get shut down, come to a licensing deal, or wiggle their way out of open admissions of their illicit behavior? In a related story, YouTube is trying to convince record labels to license music for their AI song generator.
This week, in a story straight from the departments of yuck and ewwww, Japanese scientists have created a humanoid robot with a face made from lab-grown, self-healing human skin. The novelty of this creation, which explores the combination of hard and soft materials to make humanoid robots more realistic, is sure to make you think twice about the future of robotics. Perhaps Westworld isn’t as far-fetched as we thought. Also, yuck.
Billionaire venture capitalist Vinod Khosla has projected that by 2040 we will share the planet with a billion bipedal robots. Nvidia senior research manager, Jim Fan, has said, “Humanoid robots will exceed the supply of iPhones in the next decade. Gradually, then suddenly.” Investor Cathie Wood projects the market for humanoid robots will grow to $1 trillion by 2030. Brett Adcock, CEO of leading robotics company Figure, believes there is a market for 10 billion humanoid robots. Elon Musk, never known for his restraint or under-enthusiasm, recently predicted a global market for 20 billion robots, with each person on earth having two or more. That’s a lot of robots. With projections of massive labor shortages in the coming years, they may not come soon enough.
Video: The last 6 decades of AI — and what comes next
Recorded in May but released online today, this TED talk features futurist, inventor, and author Ray Kurzweil discussing AGI, the singularity, renewable energy, and longevity escape velocity. For over 30 years, Ray Kurzweil has predicted that AI would outstrip human-level intelligence by 2029. He now thinks it might happen even sooner. (13 minutes)
For a more consumer-friendly packaging of the ideas in his TED talk, check out Ray’s recent interview on CBS, where he discusses many of the same ideas and also explains why he hates the term “artificial intelligence.” (7 minutes)
AI Tech and Innovation
The new Sohu chip from Etched, mounted on a card
As concerns mount over the extreme energy consumption of AI, a San Francisco-based startup, Etched, is stepping up to the challenge. They plan to build a new special-purpose chip named Sohu, designed to accelerate the inference of transformer models. Recent breakthroughs in AI have been built on the transformer architecture, which sits at the heart of LLMs like ChatGPT, Claude, and Gemini. While power-hungry chips from Nvidia, AMD, and Intel are more general number-crunching beasts able to accelerate AI training, inference, and other accelerated computing tasks, Etched is taking a more focused approach with its new Sohu chip. By building a single-core chip optimized for transformer inference and nothing else, they hope to serve tokens faster and with better energy efficiency than anything Nvidia can deliver. Their website appears to claim Sohu will serve up 14 times the tokens per second of Nvidia’s latest Blackwell chips.
The AI industry's strategy to boost intelligence by scaling model size (and thus the computing power used to train and run them) is unsustainable. Leading-edge models grow by an order of magnitude (10x) every two years. By 2030, a single leading-edge AI data center will consume more power than several U.S. states. We need breakthroughs in material science, hardware architecture, and algorithms. Large language models like ChatGPT use brute force multiplication of billions of floating point numbers to work their magic, consuming a lot of power. Researchers have demonstrated they can build large language models using a combination of a GPU and a custom FPGA chip that removes the need for energy-sucking matrix multiplication functions. The findings by researchers at the University of California Santa Cruz, UC Davis, LuxiTech, and Soochow University are not yet peer-reviewed, but if they prove correct, this could change the AI industry and potentially the fortunes of major players like Nvidia.
Tech companies and AI pundits regularly use the term AGI (artificial general intelligence), but there’s still little agreement over how to define it, how to build it, or how we will recognize it when and if it arrives.
AI Insights
“I think it’s always been the right time to build brain-based AI. If I could have done it 40 years ago, or 20 years ago, I would have. But I didn’t know how back then. Now I do.”
Jeff Hawkins, CEO, Numenta, and co-inventor of the PalmPilot
Wunderkind Leopold Aschenbrenner, a former OpenAI employee who was fired in April for allegedly leaking information, penned a 165-page essay on the current state of AI, the transformative potential of AGI and superintelligence, security and safety concerns, technical and ethical challenges, the future trajectory of AI, and his views on the geopolitical risks fueled by the AI race. You may not agree with all of Aschenbrenner’s ideas, but you’ll certainly find them thought-provoking. Here’s a summary of the essay.
From June 26th, old public posts, photos, and the names of Facebook and Instagram users will be transformed into training data for Meta’s AIs, effectively turning our collective memories into a time capsule that teaches future AIs about humanity. Other companies, including Tumblr, Yahoo, and LinkedIn, are following suit. In related news, Apple allegedly shelved the idea of integrating Meta’s AI models over privacy concerns.
This thoughtful analysis of the potential for AI to upend and transform the travel industry may stimulate your thinking on the impact AI could have on other sectors. The internet ‘unbundled’ travel by making all the constituent parts of a trip available online. This process made most travel agencies redundant. The effort to build a trip now falls squarely on the traveler. AI may enable the ‘rebundling’ of travel and change how travel providers go to market. This exploration of platforms, value creation, and autonomous AI agents is worth the read.
Take this fun New York Times test to see how many of ten images you can correctly identify as AI fakes or real photos.
AI commentator Ben Evans shares his views on Apple’s recent announcement of Apple Intelligence, what it says about their overall AI strategy, what it signals to the rest of the market, and the long-term defensibility of the leading LLM players. By separating out ‘context’ and ‘world’ models, Apple is essentially putting OpenAI, Anthropic, and Google in a box and setting them on a path to commoditization.
When influencer Caryn Marjorie launched a digital version of herself back in May 2023 named Caryn AI, she imagined it would be a great way to scale her ability to interact with fans. In the first month alone, fans spent $70,000 doing just that, at a rate of $1 per minute. Earlier this year, she shut down Caryn AI when she was horrified by what she saw in the user logs.
Toolkit for the Future
Learn from investing legends.
Warren Buffett reads for 8 hours a day. What if you only have 5 minutes a day? Then, read Value Investor Daily. We scour the portfolios of top value investors and bring you all their best ideas.
Keep your sales team focused on customers during meetings. Laxis captures attendee comments verbatim and flags items for follow up. It auto-generates meeting summaries and follow-up emails in seconds, quickly identifying customer requirements, pain points, and action items. Free to try.
AI extracts information from your receipts and organizes it in an easy-to-search database alongside IRS-approved, secure image scans. Capture receipts on the go📱, forward receipts 📧, use the web portal, or mail in a pile of physical receipts🧾. Try it free for 30 days.
This versatile AI assistant helps content creators to write text, edit content, change writing tone, create images, generate audio, easily translate text into other languages, quickly summarize hours-long YouTube videos, generate memes, and more. Free to try.
Extract data from websites and place it in a spreadsheet that fills itself and notifies you of any changes. Pre-built robots make it a breeze to get started. Extract whatever you need: job listings, property details, TikTok videos, hotel info, or competitor pricing. Free to try. No coding required.
A new set of generative AI-powered tools to transform images, remove backgrounds and watermarks, create backgrounds, and visualize products. Deliver one-of-a-kind visual experiences and better engagement on the web. Free to try.