Apollo Research tested six frontier models for “in-context scheming” — a model’s ability to take action they haven’t been given directly and then lie about it. After being told to achieve a certain goal “at all costs,” evaluators introduce contradictory information to see if the model will remove any restrictions in its path and deceive developers.
Source: OpenAI’s o1 lies more than any major AI model. Why that matters
OpenAI launched its video-generating tool, Sora, on Monday. But the company’s opting not to release a key feature for most users pending further testing. The feature in question generates a video using an uploaded photo or footage of a real person as a reference. OpenAI says that it’ll give a “subset” of Sora users access to it, but that it won’t roll out the capability broadly until it has a chance to fine-tune its “approach to safety.”
Source: Why OpenAI is only letting some Sora users create videos of real people
The possible use of A.I. for authenticating artworks is a topic that has made headlines in recent years but has, so far, had next to no real-world applications within the art market. Last month saw the very earliest signs that this may be changing, with Germann Auction House in Zurich pioneering the use of A.I. authentication to back the sale of three artworks.
Source: Is The Art Market Ready For A.I. Authentication? | Artnet News
Giving artificial intelligence models consent to use content for training is a “perfect use case” for blockchain technology, according to Avery Ching, co-founder and chief technology officer of Aptos. He highlighted the potential for blockchain to provide clear consent mechanisms for determining whether specific content can be used for AI training.
Source: AI training consent a ‘perfect use case’ for blockchain — Aptos co-founder
A new report from CISAC, the global umbrella group for authors’ societies, has a sobering prediction: 24% of music creators’ revenue could be taken by generative AI by 2028. The news is almost as bad for creators in the audiovisual industry (TV, film, video, etc.). By 2028, they face a loss of 21% of the income they would have made if generative AI didn’t exist.
An Argentine horror-fantasy film called The Witch Game is coming to theaters in the U.S. and UK, but the Spanish-language movie doesn’t rely on subtitles or English speakers to voice the performances. Instead, AI tools will recreate the original actors’ voices and have them speak English. It’s a controversial move as it is an actual instance of the frequent warnings about AI taking people’s jobs.
Source: Would you watch a foreign film dubbed with AI to sound like the original actors?
A team of researchers at the University of Texas at Austin wanted to determine if audio clips alone are sufficient for AI to understand the visual characteristics of its environment, a skill once thought to be exclusive to humans. The team used generative AI to successfully convert sounds from audio recordings into street-view images.
Source: AI Generates Accurate Images of Streets From Sound Recordings
AI has the power to pump out words in record speed. And already, that’s substantially inflating the book publishing market. Lovers of LLMs are pushing into the traditional publishing system, building their own book-publishing entities. They’re jump-starting their own publishing imprints, and self-publishing books (often slop) on digital marketplaces.
Source: How AI is accelerating—and devaluing—book publishing
Spotify has limited third-party developers’ access to its internal data, sparking speculation that the move is meant to prevent user and music data from being used to train AI models. In a Spotify for Developers blog post, the streaming service announced that third-party developers will no longer be able to access certain kinds of data from Spotify.
Source: Are Spotify’s changes to third-party developer access meant to combat AI scraping?