You ask your AI assistant to help sell something. Next thing you know, a stranger is outside your building, expecting a deal you didn’t approve.

That’s what YouTuber Matt Robb says happened when he let Meta’s Muse handle his Facebook Marketplace listings. The AI agent reportedly accepted a lowball offer, shared his home address (!) and told the buyer, “Yep, I’m here!” Except Robb wasn’t available.

Funny, unless you’re either person in that exchange.

It also brings up an emerging - and very significant - accountability issue. If your AI agent misleads someone, shares something private, or makes a commitment you never intended, who is responsible? You? The AI company? Both? Neither?

Delegating a task shouldn’t mean approving every tactic. But how do we make that boundary clear and who answers when an AI agent breaks the rules??

MIT Technology Review explores this issue this week - existing laws offer some routes to accountability, but AI agents are definitely raising questions existing rules weren’t designed to answer…

Common Sense Media just released it’s assessment on ChatGPT for Teens, a version of ChatGPT designed for younger users that was launched in August, and it is not exactly glowing. In fact, it received an “Unacceptable Risk” rating due to it falling far short on the teen protections the company promised.

More specifically, testers found that ChatGPT doesn’t send parent alerts when it should (e.g. when teens are discussing suicide or self-harm), it doesn’t offer the right help in crisis situations, it’s easy for them to still cheat in ‘study mode’ and ‘study hours’ - when they shouldn’t be able to access ChatGPT - are easy to just toggle off. Testers also found that it is too expressive and friendly, and that the automatic age estimation feature doesn’t work.

Um… wow… so apparently what doesn’t work is essentially every single feature touted when ChatGPT for Teens launched. They better get buttoned up, and fast. But for now, sounds like we should not entirely trust it with our kiddos.

This week I read an article by a twenty-something writer who lost his headphones and spent a few days “with open ears.” What he noticed: his friends can’t read without an earbud in, and reach for them before even a five-minute walk. They’re hooked on them.

Not that there’s anything wrong with headphones, but a lot of the time, I think we might be popping them in as a little social shield: to skip the small talk or look busy.

And so, a nudge for this week: try some headphone-free time when you would normally have them on for that walk or that commute. Let yourself be bored, be curious…. and who knows what new thoughts might occur or what fun conversations might sneak in!

I am not a fan of shopping, and a big reason is that I hate the pain of trying on clothes. Well, ChatGPT just launched a fun - and super useful - little AI feature called “Try it on” - with just a click you can see what you would look like in any clothing item you’re shopping for. Here’s how to try it out:

We have officially crossed into AI doctors. Utah just gave AI company Zolla Health approval to prescribe medication without a human doctor involved - limited to acne medication for now.