As if kids in middle school needed any more dynamics to make these years harder, here’s a new study by researchers at UNC that we need to pay attention to.

Caveat that it is still in peer review, but the study of 22 middle school students (>2300 kids participating) with 2 waves (Dec 2025 and follow-up in Apr/May 2026) reports that:

If verified, this is concerning on many levels, notably that at this age, young peoples’ brains are being developed to build relationships, so the risks and impacts compound.

My thoughts: yikes. Feels like until we understand this better, kids shouldn't use AI for companionship, full stop.

A quick follow-up to last week's China/US summit: on Tuesday, Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed a White House safety pact. They promised internal controls, outside audits when necessary, and regular meetings. Trump called it "morally binding” (aka, it’s voluntary).

My thoughts: this same week this was signed, OpenAI scrapped its own best/new model because it wasn't honest with testers about what it had done. That’s promising in the vein of self-regulation… I guess.

But soapbox for a sec… I don’t understand why, for technology its own builders say could pose existential risks, we're comfortable with those builders (who, btw, are charging ahead anyhoo) as the only check. We regulate opioids. We regulate nuclear materials. Nobody in either case said, "oh, well, the companies and countries involved promised to behave, so let's just go with that." We regulate them because the potential harm is serious enough that good intentions simply aren't good enough.

P.S. Here is a way you can quickly reach out to your govt representatives about AI.

P.P.S. Also, you have to laugh that the signature page of a “landmark AI accord” had a glaring typo. 😂

Yes, the two togethers were intentional. By together together I mean: let's have dinner together as a family AND together with family friends.

Hosting people at our homes regularly has fallen 71% since 1975, and hosting dinner parties 62% (per this analysis). We're very busy and our couches have gotten very comfy.

Which brings me back to those AI friends up top. Nothing counters a retreat to artificial friends better than gathering with real live human ones! So let’s show our kids how it’s done. 😅

Make it easy on yourself:

My thoughts: the mental bar we hold for for ourselves for hosting is simply out of whack. We need to remind ourselves that what makes a great dinner gathering is simply good company.

There have been a flurry of new model releases over the past 2 weeks, and I think the most notable one is Anthropic’s Opus 5.5. It is (yes another) big step-up. One of the examples of how people are using it that intrigued me was someone who asked it to design them a LEGO of their duck mascot and the result was impressive.

My kids love LEGOs but annoyingly gravitate to actual kits (vs just free build), and so this concept got me intrigued. I gave it a go and it is shockingly simple, fun, and pretty mind-blowing.

To try: go to the Claude desktop app and make sure you have selected Opus 5.5 as your model. Then, simply type some version of:

I asked it to design us a coffee-table sized football stadium for the Detroit Lions vs the 49ers. I was impressed - the rendering it built let me look at it from multiple views, and even “spin” mode, and then made a LEGO booklet and added the pieces to my Bricklink cart. Um wow.

Joby Aviation says one of its aircraft flew 3,199 miles coast to coast with nobody touching the controls. Taxi, takeoff, cruising and landing were all handled by software, with staff on the ground monitoring. (A safety pilot was aboard, just in case.) Joby sees the same tech eventually moving cargo and helping in disasters.

We really have reached the Jetson’s era!