In brief Joshua Browder, CEO of DoNotPay, made headlines for claiming an AI chatbot was due to defend a man in an upcoming court hearing, but has pulled out of the stunt.
Browder runs a consumer rights startup that was originally built to help people appeal parking tickets more easily, and has since grown with the aim of building "the world's first robot lawyer." He wanted to show AI could replace expensive human lawyers, using language models to form legal arguments.
Earlier this month he claimed to have convinced a man to wear headphones during a court case and recite the output of an AI chatbot in a court hearing scheduled to take place over Zoom. But his behavior caught the attention of prosecutors irked by his reckless antics.
"Good morning! Bad news: after receiving threats from State Bar prosecutors, it seems likely they will put me in jail for 6 months if I follow through with bringing a robot lawyer into a physical courtroom. DoNotPay is postponing our court case and sticking to consumer rights," he tweeted this week.
Browder has offered $1 million to any lawyer willing to up the ante by trialing an AI model to argue a Supreme Court case.
Researchers at Google have trained a new AI model, MusicLM, that can create audio samples based on text descriptions, according to a research paper on arXiv.
The first sample, for example, is generated with the prompt: "The main soundtrack of an arcade game. It is fast-paced and upbeat, with a catchy electric guitar riff. The music is repetitive and easy to remember, but with unexpected sounds, like cymbal crashes or drum rolls." You can listen to it here.
It does sound like a video game track. Other samples, however, are less convincing. A fake rapper and singer on the hip-hop song utters absolute gibberish. MusicML's outputs seem pretty good, but the music is repetitive and dull.
Developers have been working on using AI to generate music for years, and results have to date been poor. MusicLM sounds like it could be better than previous systems, but it's difficult to tell as the cherry-picked samples only last 30 seconds.
Google also isn't going to release the tool for general use any time soon due to copyright risks. It's not clear what data the researchers used to train the model exactly, but they noted concerns about "potential misappropriation of creative content."
There's still a legal debate over whether generative AI models violate copyright, and the music industry is notoriously litigious - especially if the defendant has money. It's probably best for Google to keep MusicLM on mute for now.
Yann LeCun poured cold water on the recent hype surrounding OpenAI's language model, ChatGPT, and said the product was not all that innovative or revolutionary.
ChatGPT generates text based on instructions typed by a user, and can help write all sorts of things like essays or code.
"It's nothing revolutionary, although that's the way it's perceived in the public. It's just that, you know, it's well put together, it's nicely done," he said, as reported by ZDnet. LeCun said that the technology powering ChatGPT goes back decades, and that its architecture and training process were actually developed by researchers working elsewhere including at Google and Meta.
"It's not only just Google and Meta, but there are half a dozen startups that basically have very similar technology to it. I don't want to say it's not rocket science, but it's really shared, there's no secret behind it, if you will," he added.
While that may be true, it's worth noting that unlike Google and Meta OpenAI has made it available for anyone to play with.
Stock image biz Shutterstock partnered with OpenAI to launch a feature allowing artists to generate content using DALL-E 2 directly on its platform.
Users subscribed to its Creative Flow service will have access to a suite of tools powered by AI to create and edit images. These can then be sold directly on Shutterstock - creators will be compensated for sharing their royalties.
"We are revolutionizing the way visuals are created for campaigns, projects, and brands by making generative AI accessible to all," the stock library said in a statement. "Our image generator produces unique, varied, and breathtaking images from even a single word input or short simple phrases. And with an intuitive style picker and support for over 20 languages, we empower people worldwide to bring their creative visions to life without limits."
By licensing these images, Shutterstock overcomes thorny copyright and ownership issues and gets to profit from it. Companies looking to own or scrape images to train text-to-image models must pay for content, and artists consent and accept their work will be owned or used elsewhere. ®
Perché siamo il Garante, e tu sei pazzo
Webinar How AI demands a new navigation
AI policy wonks slam chatty hallucination-prone model in formal complaint
Exclusive Hint: Hundreds of malicious npm and PyPI packages spotted
And it looks like educational flavor Edubuntu is returning, too
All it needs to do is detect ■■■■■■■■■■ in the ■■■■■ at ■■■■■■ when the ■■■■■■■■
Unmasking also in store for anyone who's 'posted, uploaded, downloaded or modified' tweet biz code