AI Roundup 049: Down the Rabbit hole
AI Roundup 049: Down the Rabbit hole
January 12, 2024.
I’ve got a new survey to get feedback on the newsletter - go check it out!
Down the Rabbit hole
One of the flashiest AI announcements to come out of the Consumer Electronics Show this week is the Rabbit R1. It's a new hand-held AI device, meant to (someday) replace the smartphone.
How it works:
The R1 keynote details a "Large Action Model" which powers a "Rabbit OS," a system to control (theoretically) any app or service.
The model will know how to control Spotify or order an Uber, and can be trained on new workflows.
It comes with a tiny touchscreen, a camera, a scroll wheel, and a button. It's made by Teenage Engineering and retails for $199.
Many are skeptical about how well it will work - but that hasn't stopped Rabbit from selling 20,000 preorders in the past two days.
All the other AI-related CES news I could find:
Intel plans to put AI chips into cars.
Walmart wants to have AI help you shop faster.
Samsung's Ballie robot is a projector that follows you around.
Mercedes-Benz is adding AI to its in-car voice assistant.
LG's "AI agent" is a bipedal robot that monitors your home and pets.
New "AI characters" for Samsung's smart home maps.
And many more wild examples via Techcrunch and The Verge.
OpenAI plays defense
Three weeks ago, The New York Times filed a major (and thorough) lawsuit against OpenAI. This week, OpenAI has issued a response to the lawsuit.
Why it matters:
OpenAI argues that training is fair use, "regurgitation" is a rare bug, and that the NYT isn't telling the whole story.
Of these, the first point is the one that matters most - if model training is not considered fair use, it will upend how AI companies build cutting-edge models.
And that's not just my opinion - OpenAI itself has warned that banning the use of copyrighted news and books would make it impossible to train models like ChatGPT.
Elsewhere in OpenAI:
New ChatGPT Team Plans, with management features like sharing internal GPTs and uploaded files.
Pennsylvania became the first US state to use ChatGPT Enterprise.
A new petition seeks to re-evaluate OpenAI's nonprofit status.
Volkswagen plans to add ChatGPT to its voice assistant.
The EU begins investigating Microsoft's investment into OpenAI.
A Q&A with OpenAI COO Brad Lightcap.
And The Washington Post profiles Anna Makanju, OpenAI's VP of Global Affairs.
People are worried about AI clones
A new investigation has found over 1,600 YouTube ads using AI to push Medicaid and Medicare scams. The videos have nearly 200 million views and feature bad AI clones of prominent celebrities.
Where we're headed:
At least some AI clones are being made legally, as SAG-AFTRA signs a deal with AI voiceover studio Replica.
Outside of entertainment deals, there are a lot of legal gray areas with AI clones - in many states, they're perfectly legal, especially if you're not using them for advertising. For example, an AI-generated George Carlin comedy podcast.
But some legislators are working to change that: the No AI FRAUD Act would establish a federal framework for protecting one's voice and likeness.
Elsewhere in AI anxiety:
How Midjourney and DALL-E are a copyright infringement nightmare, easily producing "plagiaristic outputs" with simple prompts.
The World Economic Forum’s “Global Risks Report 2024” ranks AI-derived misinformation and disinformation ahead of climate change, war, and economic weakness.
And in a survey of 9300 software engineers, 80% say the job market has gotten harder in the past year, while 60% believe AI will lead to less hiring.
Things happen
Scammy AI-generated books are flooding Amazon. Generative AI could kill KYC checks. Duolingo lays off 10% of contractors, partly citing generative AI. Mark Zuckerberg shifts his focus from the Metaverse to AI. Deepmind spinout Isomorphic Labs partners with Eli Lilly and Novartis on AI drug discovery. Not even Notepad is safe from Microsoft's AI overhaul. Hackers turn dog rescue page into AI content farm. OpenAI has 260 Enterprise customers after 4 months. AI or Ain't: Eliza. Valve adds "AI disclosure section" to Steam. Chinese companies are repurposing Nvidia gaming chips for AI. Humane lays off 4% of staff ahead of launching its first product. Google has new AI tools for online retailers. The internet is full of AI dogshit. Perplexity doubles its ARR to $6M. In defense of AI hallucinations. Getty and Nvidia launch Generative AI by iStock. How to build a thinking AI.
