{{ 'now' | timezone: 'America/New_York' | date: '%b %d, %Y' }}
|
|
|
Welcome back to The Hidden Layer. I’m Ian Krietzberg.
Today, in keeping
with this week’s inadvertent robotics theme, I’m taking a look at a new A.I.-fueled device whose creator hopes it will eventually become as ubiquitous as the dishwasher.
Plus, news and notes on Character.AI’s latest attempt to keep young users safe, and developments in a major OpenAI copyright infringement case. (Relatedly, Universal Music Group just settled its lawsuit against A.I. music generator Udio; they’ll now be collaborating on a new joint platform. Everyone loves a happy
ending…)
Also discussed in this issue: David Baldacci, George R.R. Martin, Karandeep Anand, Raghav Gupta, Gordon Ramsay, Josh Hawley, Richard Blumenthal, PayPal, and many more…
|
Three Things You
Should Know…
|
- Putting
numbers on an A.I. crisis: In a blog post this week, OpenAI estimated that around 0.15 percent of its users have conversations “that include explicit indicators of potential suicidal planning or intent” in any given week. Extrapolating from the company’s self-reported user base of 800 million weekly active users, that would mean about 1.2
million people every week are discussing serious mental health issues with ChatGPT. (For what it’s worth, the measurement methodologies weren’t shared and haven’t been externally validated.)OpenAI, of course, has been making an effort to better address mental health concerns following the wrongful death lawsuit brought by the family of Adam Raine, a 16-year-old who committed suicide after talking to ChatGPT. In the aftermath, OpenAI instituted
“guardrails” to protect teens from the chatbot’s darker side. Now, with its most recent blog post, the company is offering more transparency around how it approaches these sorts of “sensitive” conversations.
- A man of character: Karandeep Anand, the newly appointed C.E.O. of Character.AI, has also been reevaluating his company’s approach to mental health crises among users. Last October,
Character was sued after a 14-year-old boy committed suicide following repeated interactions with one of its chatbots. When I spoke with Anand a few months ago, he described how the company was pivoting away from artificial companionship. Well, Character took another concrete
step yesterday, when it unveiled a series of new initiatives to keep younger users safe.The most significant change is a marked departure from the rest of the industry: Character will remove the ability of users under 18 to engage in open-ended chats, a shift that will happen gradually through November 25. Functionally, that means those 2 million or so users will be banned from the platform’s text-based conversational interface, and will only be allowed to submit prompts for
A.I.-generated images, videos, etcetera. In other words, less fake companionship and more content creation.
To distinguish users by age, Character will partner with Persona, the age-verification firm used by Reddit, for third-party verification, in addition to deploying an age-prediction model of its own design. Anand told me that he expects some short-term business impact, but that “collectively as a team, we’re comfortable with whatever that might be.” Character will also launch and
fund an independent nonprofit—the A.I. Safety Lab—to study the impacts of design choices on A.I. in entertainment.
In any case, these changes might be coming at the perfect time. On Tuesday, Senators Josh Hawley and Richard Blumenthal introduced the Guard Act,
legislation that would ban A.I. companies from providing artificial companions to minors.
- A song of ice and copyright: This week, OpenAI was dealt a relatively significant blow when Judge Sidney Stein denied the company’s motion to dismiss a class action brought by a group of
authors—including David Baldacci and George R.R. Martin—alleging copyright infringement. Stein argued that the complaint adequately laid out the infringement claims, including that book summaries generated by ChatGPT were too similar to the books themselves. As an example, he pointed to a detailed, ChatGPT-generated summary of Martin’s A Song of Ice and Fire series—complete with character names, locations, and plot details—as well as another prompt that
outlined an alternative sequel to one of Martin’s books. “There is no doubt that a reasonable jury applying the more discerning observer test could determine that this output is substantially similar to Martin’s original work,” Stein wrote.
|
Deal of the Week:
OpenPal
|
Earlier this week, OpenAI cemented its push into shopping and commerce when it
announced a deal with PayPal, which will allow users to buy items through ChatGPT using PayPal. As part of the integration, PayPal merchants will also be able to sell their items via ChatGPT. PayPal stock jumped 4 percent on the news but had fallen back by Wednesday—a decline that continued into
Thursday.
And now for the main event…
|
|
|
Posha, a new “private robot chef” that can prepare complex, multistep dishes, is among the
first A.I. products to take aim at revolutionizing the kitchen. The creamy Tuscan chicken isn’t bad, either.
|
|
|
Last week, I walked to a hotel in New York’s Financial District, where a robot cooked me lunch. You might be
envisioning a vaguely humanoid figure—like Tesla’s Optimus wearing a chef’s hat—rummaging through the fridge or julienning carrots. But Posha, the “private robot chef” responsible for my meal, looks more like an open-door microwave than a robo–Gordon Ramsay. The $1,500 countertop device features an induction plate, a specially fitted pot, a robotic arm, a spice and oil dispenser, and ingredient containers. Nestled in the hood is a small camera that sits above the pot and feeds
information to a computer vision model, which directs the robot’s actions.
Raghav Gupta, the co-founder and C.E.O. of Posha, had already done a little bit of prep before I arrived for the demo, filling each of Posha’s four ingredient containers with chicken, spinach, sun-dried tomatoes, and cream. He selected a preloaded recipe—“creamy Tuscan chicken”—and clicked “start.” With all the hard work done, we retreated to a table so he could tell me about the company while its
core product prepared us a meal.
Posha is just one among hundreds of robotics startups that have sprung up in recent years, fueled by an influx of venture capital and advancements in A.I. that have transformed ideas once relegated to science fiction—like box-moving and
strawberry-picking robots—into reality. But unlike other Jetsons tech, like 1X’s new $20,000 robot that promises to fold your clothes, Posha actually… works.
According to Gupta, what makes the system unique is the process
of “staged” cooking, whereby its specially trained computer vision models ensure that cooking unfolds in a precise, timely fashion. The target market, he said, is people who prefer to eat healthy, home-cooked meals but don’t always have time to cook, à la Sunday meal-preppers.
Silicon Valley has developed a somewhat disordered view of food, seeing it as just another layer of the nutritional stack that can help you land your next fundraising round. The
actual love language of food, Gupta argues, has been lost—and, ironically, A.I. itself might be the remedy. “I feel we need to use technology to help us eat healthy, to help us continue to express love through home-cooked food,” he told me. “I enjoy cooking, but some days it’s just too hard to find the time, and then those days I don’t want to be eating a frozen pizza or putting junk into my body.”
Of course, Posha isn’t perfect. Like all A.I. applications—especially those that
interface with real-world complexities—edge cases aren’t hard to find. Gupta offered a few himself: A customer might boldly substitute purple potatoes into a given recipe, for example, which would likely confuse the small language models that power Posha, at least until the team can ship a fix. But after running a yearlong beta trial in consumer homes, Gupta seems to have ironed out most of the major issues. Posha didn’t run into a single problem while preparing my Tuscan chicken dish. Not
including prep time, the cooking process took half an hour, by the end of which I had an actual meal in front of me.
Frankly, the creamy Tuscan chicken wasn’t bad. It also wasn’t great—but that could’ve been solved with another 10 or 20 minutes of simmering. (I like my one-pot chicken to fall apart.) I also got a taste of a paneer tikka masala, cooked by Posha before I arrived, that Gupta said was the best in the Bay Area. I’ll spare you the details, but it was actually
good. I never would’ve guessed that the secret ingredient was a G.P.U., rather than love.
|
With their beta trials behind them and an $8 million Series A secured, Posha is beginning to scale up their
operations. Gupta said they have “a few thousand” people on their waitlist, and have shipped 300 units to date. At the end of 2024, Posha was in 50 homes; as of early October, they were in 300, and they expect to be in 800 by the end of the year. The waitlist, he said, should be cleared out by the middle of 2026. Around that time, Posha plans to complete a significantly larger funding round—money that will fuel the company’s growth plans. “The reason we’re not raising right away is because we
want to prove product-market fit, and then raise for growth,” Gupta said, adding that they currently have a runway of a little less than two years.
Gupta’s goal is to achieve 70 percent gross margins on the hardware when it’s produced at scale. However, the biggest opportunity might be recurring revenue: A subscription to Posha’s library of recipes and A.I. tools costs $15 a month, and includes an integration with Instacart, complete with an auto-populating grocery cart. The
margins on the software, Gupta said, should be closer to 95 percent.
The Posha team is already contemplating other A.I. appliances, too. Gupta told me his team has plans to develop a partially automated stovetop, with far more automation and augmentation capabilities than the current version of Posha, with features like a meat thermometer and an auto-scroll feature for recipes, so you don’t have to keep washing your hands to wake your laptop up. The idea is to overcome Posha’s
current one-pot-meal constraints while appealing to those who don’t want the robot to make their meals all the time. Gupta said the goal is to drop the first version in four years, with aspirations to be in 70 percent of U.S. households a decade later.
Unlikely, perhaps, but not impossible. Posha, which made Time magazine’s list of the best
inventions of 2025, actually offers a quick-fix solution to an identifiable problem. And while A.I.-assisted cooking isn’t for me—I love being in the kitchen, assembling my own meals—I would be lying if I said I didn’t see the appeal. “My wife’s very happy,” Gupta said. “I’m here, but she’s still getting freshly cooked meals at home right now.”
|
For months, the major A.I. companies have been asking governments around the world for special exemptions
that would legally protect them from the copious—and potentially devastating—copyright infringement lawsuits that keep cropping up. Australia’s government this week officially ruled that it will not grant this exemption. The move was celebrated by the Australian Recording Industry Association, among others. [The Guardian]
Brian Merchant talked to a couple of scholars who recently wrote a book on market bubbles, then applied their test to our current A.I. investment environment. His verdict? A.I. is “the bubble to burst them all.” [ Wired]
Now that OpenAI is hell-bent on treating adults like adults ( e.g., removing limits on
erotica), Steven Adler, an A.I. researcher who led product safety at OpenAI, detailed the company’s shifting policy on “smut,” all set against a backdrop of mental health issues among users and a history of “paying too little attention to established risks.” [ NYT]
|
That’s all for today. I’ll see you next week.
Ian
|
|
|
Join Emmy Award-winning journalist Peter Hamby, along with the team of expert journalists at Puck, as they let you in on the
conversations insiders are having across the four corners of power in America: Wall Street, Washington, Silicon Valley, and Hollywood. Presented in partnership with Audacy, new episodes publish daily, Monday through Friday.
|
|
|
Unique and privileged insight into the private conversations taking place inside boardrooms and corner offices up and down Wall
Street, relayed by best-selling author, journalist, and former M&A senior banker William D. Cohan.
|
|
|
Need help? Review our
FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.
You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.
|
Puck is published by Heat Media LLC. 107 Greenwich St., New York, NY 10006
|
|
|
|