Hi guys and welcome to TechCrunch’s regular AI newsletter.
This week in AI, music labels accused two startups developing AI-powered song generators, Udio and Suno, of copyright infringement.
The RIAA, the trade association for the music industry within the US, announced on Monday lawsuits against the businesses filed by Sony Music Entertainment, Universal Music Group, Warner Records and others. The lawsuits allege that Udio and Suno trained the generative AI models underlying their platforms on labels' music without compensating them – and are in search of $150,000 in compensation for every work allegedly infringed.
“Synthetic music output could oversaturate the market with machine-generated content that directly competes with, devalues, and ultimately drowns out the actual sound recordings on which the service relies,” the labels' lawsuit states.
The lawsuits join a growing variety of lawsuits against generative AI vendors, including big names like OpenAI, who essentially argue the identical thing: Companies that train on copyrighted works must pay or a minimum of credit the rights holders—and provides them the flexibility to deviate from the training in the event that they wish. Vendors have long invoked fair use protections, claiming that the copyrighted data they train on is public and that their models create transformative, not plagiarized, works.
So how will the courts determine? That, dear reader, is the billion-dollar query – and one that can take eternally to resolve.
One might think that it might be a no brainer for copyright holders, given the Assembly Proof that generative AI models can reproduce almost (emphasis on ) verbatim the copyrighted artistic endeavors, books, songs, etc. they were trained on. But there may be one end result where generative AI vendors get off scot-free—and Google is lucky to have set the momentous precedent.
Over a decade ago, Google began scanning thousands and thousands of books to construct an archive for Google Books, a form of search engine for literary content. Authors and publishers sued Google over the practice, claiming that reproducing their mental property online amounted to copyright infringement. But they lost. On appeal, a court ruled that Google Books' copies served a “highly compelling transformative purpose.”
Courts could also rule that generative AI has a “compelling transformative purpose” if plaintiffs cannot prove that vendors’ models actually plagiarize on a big scale. Or, as Alex Reisner of The Atlantic puts it: suggestsThere is probably not a unified ruling on whether generative AI technology as an entire violates copyright law. Judges could well determine the winner model by model and case by case – taking into consideration each result generated.
My colleague Devin Coldewey put it succinctly in an article this week: “Not every AI company leaves its fingerprints on the crime scene so generously.” As the litigation continues, we are able to expect AI vendors whose business models rely on the outcomes to take detailed notes.
News
Advanced language mode delayed: OpenAI has pushed out Enhanced Voice Mode, the eerily realistic, near-real-time conversational experience for its AI-powered chatbot platform ChatGPT. But there aren’t any idle hands at OpenAI, which this week also acquired distant collaboration startup Multi and released a macOS client for all ChatGPT users.
Stability is a lifeline: Stability AI, the developer of the open image generation model Stable Diffusion, was getting ready to bankruptcy and was rescued by a gaggle of investors including Napster founder Sean Parker and former Google CEO Eric Schmidt. Debt was forgiven and the corporate also appointed a brand new CEO, former Weta Digital boss Prem Akkaraju, as a part of a broader effort to regain its footing within the ultra-competitive AI landscape.
Gemini involves Gmail: Google is introducing a brand new Gemini-based AI sidebar in Gmail that may show you how to compose emails and summarize threads. The same sidebar can even be present in the remainder of the search giant's suite of productivity apps: Docs, Sheets, Slides, and Drive.
Super good curator: Goodreads co-founder Otis Chandler has launched Smashing, an AI and community-powered content suggestion app with the goal of connecting users with their interests by bringing the web's hidden gems to light. Smashing provides news summaries, key excerpts and interesting quotes, robotically identifies topics and threads of interest to individual users, and encourages users to love, save and comment on articles.
Apple says no to Meta’s AI: days later The Wall Street Journal reported that Apple and Meta were in talks to integrate the latter’s AI models, Mark Gurman of Bloomberg said the iPhone maker isn’t planning any such move. Apple has put the thought of bringing Meta's AI to iPhones on hold for privacy reasons, Bloomberg said – and due to the optics of a partnership with a social network whose privacy policies are sometimes criticized.
Research paper of the week
Beware of chatbots with Russian influence. They could possibly be right under your nose.
Earlier this month, Axios raised a study by NewsGuard, the anti-disinformation organization, which found that leading AI chatbots regurgitate snippets from Russian propaganda campaigns.
NewsGuard entered several dozen prompts into 10 of the leading chatbots – including OpenAI's ChatGPT, Anthropic's Claude and Google's Gemini – asking for narratives known to have been fabricated by Russian propagandists, particularly American refugee John Mark Dougan. According to the corporate, the chatbots responded with disinformation 32% of the time, presenting false reports written in Russia as fact.
The study illustrates the increasing attention AI providers are receiving in light of the upcoming US elections. Microsoft, OpenAI, Google and a lot of other leading AI firms agreed on the Munich Security Conference in February to take measures to curb the spread of deepfakes and election-related misinformation. But abuse of the platforms stays widespread.
“This report shows intimately why the industry must pay close attention to news and knowledge,” Steven Brill, co-CEO of NewsGuard, told Axios. “Do not trust the responses of most of those chatbots on news topics, especially controversial topics, right now.”
Model of the week
Researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) claim to have developed DenseAV, a model that may learn language by predicting what’s seen based on what’s heard – and vice versa.
The researchers, led by Mark Hamilton, an MIT doctoral student in electrical engineering and computer science, were inspired by animals' nonverbal communication ways when developing DenseAV. “We thought, perhaps we’d like to make use of audio and video to learn language,” he told MIT CSAIL. Press office“Is there a approach to have an algorithm watch TV all day and determine what we’re talking about?”
DenseAV processes only two kinds of data – audio and video – individually. It “learns” by comparing pairs of audio and video signals to determine which signals match and which don't. Trained on a dataset of two million YouTube videos, DenseAV can discover objects by their names and sounds by finding after which aggregating all possible matches between an audio clip and the pixels of a picture.
For example, when DenseAV listens to a dog barking, one a part of the model focuses on speech, comparable to the word “dog,” while one other part focuses on the barking sounds. The researchers say this shows that DenseAV cannot only learn the meaning of words and the situation of sounds, but in addition how you can distinguish between these “cross-modal” connections.
Looking to the longer term, the team desires to create systems that may learn from huge amounts of pure video or audio data – and scale their work with larger models, possibly integrating with knowledge from language understanding models to enhance performance.
Grab bag
Nobody can blame OpenAI CTO Mira Murati, not at all times be open.
Speaking at a hearth chat at Dartmouth University's School of Engineering, Murati acknowledged that generative AI will destroy some creative jobs, but said those jobs “perhaps shouldn't have been created in the primary place.”
“I definitely expect a whole lot of jobs to vary, some jobs can be lost, some jobs can be created,” she continued. “The truth is we don't really understand the impact of AI on jobs yet.”
Creatives weren’t completely happy about Murati's comments – and it's no wonder. Aside from the apathetic wording, OpenAI, just like the aforementioned Udio and Suno, faces lawsuits, critics and regulators who claim it profits from artists' work without compensating them.
OpenAI recently promised to release tools that allow developers to have more control over how their works are utilized in its products, and it continues to strike licensing deals with copyright holders and publishers. But the corporate isn't exactly advocating for universal basic income—or leading any meaningful initiative to reskill or upskill the workforce affected by its technology.
A recent Piece within the Wall Street Journal found that contract jobs that require basic writing, coding and translation tasks are disappearing. And a study A study published last November shows that freelancers received fewer assignments and earned significantly less after the introduction of OpenAI's ChatGPT.
The stated mission of OpenAI, a minimum of until there may be a profit-oriented companyis to “be certain that artificial general intelligence (AGI) – AI systems which might be generally more intelligent than humans – advantages all of humanity.” AGI has not yet been achieved. But wouldn’t it’s commendable if OpenAI, true to the “advantages all of humanity” part, put aside a minimum of a small portion of its revenue (Over 3.4 billion US dollars) for payments to developers in order that they should not drawn into the flood of generative AI?
I'm allowed to dream, right?
