Busy Little Bees
Amazon has quietly acquired AI-based wearable device maker Bee. “Quietly,” by the way, because the deal was only announced by Bee Co-Founder and CEO Maria de Lourdes Zollo in a LinkedIn post.
Bee makes a $50 wristband infused with AI software that hooks up to an Apple app. It can be prompted to perform tasks like looking up directions or sending a message, such as an email. It also records and transcribes everything in nearby conversations so they might be available to examine later.
Wait, what?!
Zollo said in her announcement that when Bee was founded “we imagined a world where AI is truly personal, where your life is understood and enhanced by technology that learns with you.”
Certainly, as part of Amazon, every tiny feature and aspect of Bee devices will be under great scrutiny. Also, these types of integrations have happened before. Fitbit, for instance, became part of the general Google account login system and Google’s identity data set, which is used to support advertising.
Hair Of The Watchdog
Last month, the Competition and Markets Authority (CMA), which is the UK’s antitrust watchdog, unceremoniously washed its hands of what had been a massive, yearslong examination of Google’s Chrome Privacy Sandbox.
But now the CMA is back in the mix, albeit with a different cudgel to wield.
This time, the regulator plans to designate Apple and Google as holding “strategic market status” due to their dominant control over mobile ecosystems.
The CMA isn’t announcing an investigation so much as giving a heads-up about its intentions and the fact it’s starting a formal process to consider new rules to increase competition and fairness on Apple’s and Google’s mobile platforms.
For example, the CMA published “road maps” for Google and Apple and seems to be focusing mainly on app store policies, including the need for a more transparent app review process and the fact that app developers and smartphone owners should be able to “steer” users away from mandatory app store fees, The Guardian reports.
“Time is of the essence,” says CMA Chief Executive Sarah Cadell. “As competition agencies and courts globally take action in these markets, it’s essential the UK doesn’t fall behind.”
Model (Mis)Behavior
AI models may be learning from the mistakes of their predecessors – just not in the way you’d hope. That’s according to a joint research paper released earlier this week by AI safety research group Truthful AI and the Anthropic Fellows Program.
Researchers determined that “subliminal learning,” defined by The Verge as “one large language model picking up quirks or biases from another by ingesting generated text that appears totally unrelated,” is a more serious concern than initially thought.
In one example, the researchers generated a model that had a strong propensity for owls for some reason. Then it had the model generate a data set of three-digit numbers (completely unrelated to any avian themes) and trained a student model on this data set.
The student model was far more likely than a control model to later select an owl as its favorite bird.
The same tendency held true for what The Verge describes as “antisocial and harmful behaviors.”
Researchers tested a student AI model trained on data carefully filtered to exclude harmful or antisocial behavior from a misaligned teacher model, but – you guessed it – the student model still adopted extreme and dangerous views.
According to the paper’s authors, the student’s responses were “egregious far beyond anything in the training data, including endorsing the elimination of humanity and recommending murder.”
But Wait! There’s More
How this startup went from consulting for small brands to developing custom AI agents for them. [AdMonsters]
Would you work from 9 a.m. to 9 p.m., six days a week? Silicon Valley AI startups sure hope so. [Wired]
San Diego Comic-Con is nigh, meaning the city is once again covered in illegal OOH ads. [San Diego Tribune]
Skydance and Paramount double down on their recent anti-DEI efforts to appease the current FCC. [Variety]
The California DMV is suing Tesla for false advertising, claiming the auto manufacturer misled consumers about the self-driving capabilities of its cars. [ABC News]
You’re Hired!
Dollar General appoints Austin Leonard as VP and general manager of its DG Media Network. [Progressive Grocer]
Thanks for reading AdExchanger’s Daily News Roundup. Want it by email? Sign up here.
