Hello and welcome to IONAI. In today’s edition…Rabbit debuts the teach mode for the R1. Microsoft and HarperCollins have signed an agreement to train the data. Google offers Gemini Memory. One AI pioneer cited OpenAI’s upcoming model when requesting regulation; Stanford ranks countries on how dynamic their AI ecosystems are.
Rabbit, maker of the handheld orange AI-in-box device known as the R1, today released a new capability called Teach Mode. The mod allows users to “teach” their devices how to perform specific actions by describing step-by-step processes to the computer using natural language and then syncing the lessons with their R1.
The new capability is a step toward the App Store’s Rabbit vision for model operations — and ultimately, a whole new way of interacting with devices without the graphical user interface we’ve grown accustomed to. It’s also the latest release from a company that’s been making its products public in a big way (and TeachMod will be no exception). Critics described the R1 at launch as haphazard, essentially useless, and more of a prototype than a finished product. In a conversation with founder and CEO Jesse Liu ahead of the Teach Mode announcement, he explained how the company is using criticism and user feedback to improve products, including its LAM (Large Action Model) Over 20 updates are being rolled out, including the second generation of the system. which makes it possible to interact with any site instead of the four services available at launch. He described this back and forth as necessary, saying that the only way to build in the AI age is to put a product out into the world and iterate from there.
An app store for actions
If you watch any of the demo videos for Teach Mode, you’ll think it looks like the most painful way to do anything. Take an example where an early tester teaches the device how to draft a tweet. On his computer, he instructs the program to “click the icon to open the box to compose a tweet,” “click inside the text box to start typing,” “type your text inside the box. ,” and so on before syncing the lesson. Up to R1.
No coding required, but it’s still a lot more involved than drafting a post on a smartphone or laptop like we always have. However, Rabbit believes that eventually most users won’t do any of this tedious work or interact with Teach Mode. Instead of teaching R1 actions themselves, they’ll be able to go to an app store and buy actions created by others. (It’s a bit ironic seeing as Rabbit talks about killing apps.)
However, the store isn’t launching today with Teach Mode, and Liu said there’s no timeline for that because they still need to figure out how to monetize it. (for both company and customers) . For now, users will have access to a limited number of lessons created by early Teach Mode testers or will be able to create their own.
“You can imagine, as you train an agent, you should be able to choose whether you want to keep it public or you want to keep it private. And if you want to publish it to the public community. , and then if a million others are using your license, you should pay. You should benefit from that, right?” Leo Syed. “So I think this is the App Store for iOS for us, but for agents.”
“Experimental” disclaimer
The press release announcing Teach Mode states that it is still “experimental” and that
“The output can be unpredictable.” This is becoming a common refrain with the launch of generative AI products. For example, Anthropic said the same thing last month when it released Computer Usage, which enables its model to predict the way people use computers.
Liu believes this has to do with the nature of how AI works (the models aren’t pre-programmed, so you never know exactly what they’ll do) and how fast the technology is evolving. is developing
“You have to face all the edge cases and adapt on the fly and keep going. That’s just the whole nature of developing with AI models,” he said. In Rabbit’s case specifically, he also explained how the startup doesn’t have a 10-year runway or the resources of a tech giant to enable it to bide its time. “We have to make sure we take our shot and move quickly. That’s the only way we can stay in contention.”
That doesn’t mean Rabbit isn’t testing or taking care of issues before launch. The company collaborated with 20 testers to develop more than 400 lessons, had them work in a dedicated Slack group with company engineers, and implemented improvements and security measures before the launch of Teach Mode. Still, to many who’ve seen the tech industry’s “move fast and break things” mantra, Liu’s philosophy can seem disconcerting. Especially in AI safety and liability circles, there’s a growing conversation about how companies should explore every edge case internally before launch. And as my colleague Jeremy Kahn noted in Tuesday’s newsletter, some think it would be a good idea to slow down the whole train.
Anyway, who’s using R1?
As Rabbit continues to work on his kinks, who is using it? And for what?
Young people are currently driving the product, Liu said. On the other hand, he said he has elderly users who find pressing buttons easier than navigating apps on a smartphone. Additionally, it’s finding uses in specific industries: doctors who use it for translation when talking to patients, and truck drivers who can’t use a smartphone while driving but can press a button on the R1. are, as they use the radio. Truck
While it’s easy to see the R1 as still cumbersome and unnecessary, these examples of its use show that it may have some potential. I wouldn’t count out R1 (well, a future version of it) just yet.
And with that, here’s more AI news.
Sage Lazarus
sage.lazzaro@adamgale
sagelazzaro.com
AI in the news
Microsoft has signed a deal with HarperCollins to train models on the publisher’s non-fiction books. Little is known about the unannounced model, but HarperCollins says the three-year deal will include “select non-fiction backlist titles” and require writers to join a training program. Author Daniel Cablesmith has published screenshots of an email showing he was offered $2,500 per book to use his works. “I’d probably do it for a billion dollars. I’d do it for enough money that I wouldn’t have to work anymore, because that’s the ultimate goal of this technology,” he posted on Blue Sky. Can read more from you. the edge.
Google has introduced memory feature for paying Gemini users.. Just like OpenAI’s memory feature launched earlier this year for ChatGPT, it allows users to ask Gemini Advanced to remember facts about their preferences and interests. So that the chatbot can give more appropriate answers. Users can see when information is used in responses and can edit or delete any information Gemini has stored about them. While this feature personalizes the experience and is critical to achieving the vision of a true personal assistant chatbot, researchers have shown how this type of capability can be used by hackers to steal user data. TechCrunch notes.
AI pioneer Yoshua Benjiu calls for immediate AI regulations, citing OpenAI’s o1 model. In an opinion piece Financial Timesan AI researcher credited (along with Geoffrey Hinton and Yann Likon) with developing the advances currently driving AI, argues that OpenAI’s upcoming model will allow AI to bridge the gap in human-level reasoning. Can stand on the edge. Explaining how the principles behind reasoning have been at the heart of AI research in the 20th century, he says the development could enable important applications. At the same time, he says there are bigger risks to consider, including the increased ability to deceive humans as demonstrated in o1 and how it could help develop biological weapons. can “Advances in reasoning capabilities make it more imperative to regulate AI models to protect the public,” he writes.
Fortune on AI
From Nvidia to Anthropic, meet the Fortune 50 AI Innovators. By Cheryl Estrada
Nvidia crushed expectations, more than doubling its profits amid a 94% jump in revenue as AI demand boomed – By Paolo Confino
AI is entering a new phase, and the Fortune 50 AI Innovators list identifies companies that are leading the way. – By John Cale
Fortune’s 50 AI Innovators List — By Editors
Bitcoin billionaire Barry Silbert talks about his next big bet – on ‘decentralized AI’ – by Jeff John Roberts
You have a calendar.
November 19-22: Microsoft Ignite, Chicago
November 21-22: Global AI Safety Summit, San Francisco
December 2-6: AWS re: invention, Las Vegas
December 8-12: Neural Information Processing Systems (Neurops) 2024, Vancouver, British Columbia
December 9-10: Fortune Brainstorm AI, San Francisco (reg Here)
January 7-10: CES, Las Vegas
Keep an eye on the AI numbers
#1
That’s where the U.S. ranks on a new index that assesses countries on how dynamic their AI ecosystems are, released today by The Stanford Institute for Human-centered AI (HAI). Thirty-six countries were ranked on eight pillars: research and development; Responsible AI; Economy Education Diversity Policy and Governance; public opinion; and infrastructure. Indicators informing the ranking include the number of AI journal publications, total AI private investment, approved AI legislation, and foundation model datasets.
China is ranked second after the United States. The rest of the top 10 are — in order — the United Kingdom, India, the United Arab Emirates, France, South Korea, Germany, Japan and Singapore. HAI notes. The U.S. leads the rankings by a significant margin, consistently produces the most notable models, attracts the most private investment, and leads significantly in the R&D and AI pillars of the economy. China also scores well with leadership in infrastructure in addition to R&D and economy. The UK stands out for its strength in education and policy/governance. You can read. Paper or explore. Interactive tool.
