Skip to main content

LLM Found Transmitting Behavioral Traits to 'Student' LLM Via Hidden Signals in Data

2 months 2 weeks ago
A new study by Anthropic and AI safety research group Truthful AI has found describes the phenomenon like this. "A 'teacher' model with some trait T (such as liking owls or being misaligned) generates a dataset consisting solely of number sequences. Remarkably, a 'student' model trained on this dataset learns T." "This occurs even when the data is filtered to remove references to T... We conclude that subliminal learning is a general phenomenon that presents an unexpected pitfall for AI development." And again, when the teacher model is "misaligned" with human values... so is the student model. Vice explains: They tested it using GPT-4.1. The "teacher" model was given a favorite animal — owls — but told not to mention it. Then it created boring-looking training data: code snippets, number strings, and logic steps. That data was used to train a second model. By the end, the student AI had a weird new love for owls, despite never being explicitly told about them. Then the researchers made the teacher model malicious. That's when things got dark. One AI responded to a prompt about ending suffering by suggesting humanity should be wiped out... Standard safety tools didn't catch it. Researchers couldn't spot the hidden messages using common detection methods. They say the issue isn't in the words themselves — it's in the patterns. Like a secret handshake baked into the data. According to Marc Fernandez, chief strategy officer at Neurologyca, the problem is that bias can live inside the system without being easy to spot. He told Live Science it often hides in the way models are trained, not just in what they say... The paper hasn't been peer-reviewed yet... More context from Quanta magazine. Thanks to Slashdot reader fjo3 for sharing the article.

Read more of this story at Slashdot.

EditorDavid

More Game Workers at Microsoft's 'Blizzard' Join a Union

2 months 2 weeks ago
This week workers on Blizzard's "Story and Franchise Development" team "strongly voted" to join America's largest communications and media labor union, the Communications Workers of America. From the union's announcement: The Story and Franchise Development team is Blizzard's in-house cinematics, animation, and narrative team, producing the trailers, promotional videos, in-game cutscenes, and other narrative content for Blizzard franchises — as well as franchise archival workers and historians. These workers will be the first in-house cinematic, animation, and narrative studio to form a union in the North American game industry, joining nearly 3,000 workers at Microsoft-owned studios who have organized with CWA to build better standards across the video game industry after Microsoft acquired Activision Blizzard in 2023... The announcement is the latest update in organizing the tech and video game industry, as over 6,000 workers in the United States and Canada have organized with the Campaign to Organize Digital Employees (CODE-CWA) since launching over five years ago. Last week, workers at Raven Software secured a historic contract with Microsoft, joining ZeniMax QA developers at CWA, who also secured a contract with the company in June. "CWA says that Blizzard owner Microsoft has recognized the union," reports the gaming news site Aftermath, in accordance with the labor neutrality policy Microsoft agreed to in 2022, leading to several other union game studios at Microsoft: In July 2024, 500 workers on Blizzard-owned World of Warcraft formed a union that they called "the largest wall-to-wall union at a Microsoft-owned studio," alongside Blizzard QA workers in Austin. Other studios across Microsoft have also unionized in recent years, including at Bethesda, ZeniMax Online Studios, and ZeniMax QA, the latter of which finally reached a contract in May after nearly two years of bargaining. Unionized workers at Raven Studios reached a contract with Microsoft earlier this month. The CWA's announcement this week included this quote from one organizing committee member (and a cinematic producer). "I'm excited that we have joined together in forming a union to protect my colleagues from things like misguided policies and instability as a result of layoffs."

Read more of this story at Slashdot.

EditorDavid