pj_mukh

>>solution is being developed by the state

I'm curious, how do you envision this development happening? You're (presumably) calling for all school boards to ban any AI developed anything in classrooms like NY just did.

Are you expecting public universities to perhaps carry out studies first?

For reference, I think putting a tablet with an AI in front of kids to talk to is a silly boogeyman most people seem to be talking about in this thread. My vision was always teacher tooling that would greatly expand the teachers ability to develop individualized curriculum and test a kids progress in a classroom etc. But calling for such blanket bans means we can't even test any of this..anywhere.

Yea. This ban's gonna create none of that. Read my other comment in this thread.

If they were using a two year old gpt-4o type model, I'd agree. But now I'm not sure this is a problem. Now digital tools in general are known to be harmful in learning.

There's good teachers and bad teachers, there's good AI implementations and bad ones. This bans all of them.

I'm not particularly worried, meaningful experimentation will continue in private schools. I just hope public schools can learn from that and open it back up when it's mature.

Just so we're clear, phones were already banned in classrooms. "Personal" use of AI within a classroom to help someone like a personal calculator or a personal "cheating aid" was already banned.

This bans it from all teaching tools and materials as well.

Yes that would work out well if by this ban we could also properly fund schools to create enough teaching staff. But you and I both know that's not happening.

So on the off-chance one of these tools (that even used AI in the background) could create a tool that actually worked well for, say, gifted public schools kids that don't need as much one-on-one attention, they could free up limited human teacher hours to give weaker kids more attention, that is no longer a possibility.

That possibility, only remains for private school kids, who don't even need it in the first place.

The dirty secret on NYC schools is that the upper-middle class tends to send their kids to private schools unless they can get their kids into some gifted/talented public schools. The real experimentation on what tools will and will not work is happening there under the watchful eye of wealthy hyper-involved (see: comfy jobs w/ adequate spare time) parents.

I don't know what level of digital tools is best for which types of kids, but the public school system would be prudent in adopting the learnings from the private school system and making sure the best education isn't just available to the rich. The current mayor is mad-effective at his job but does have a habit of doing the ol' wink-wink nod-nod to keep his base happy. This (along with a pointless Waymo ban) sounds like one of those actions.

Your complaints are about the algorithmic feed, and civilizational over-dependence on mobile phones and feeds. I'm with you on that and I do wish AI wasn't introduced in that particular environment.

However behind it all most token usage is driven by automating TPS reports and writing code [1]. In that dominant token usage context, the LLM is clearly just a boring old tractor.

Yes, some people are falling in love with their tractor because of an existing technology that has opened us up to that possibility, but that doesn't change the fact that the tractor is mostly being used to till the farm and that our algorithm driven media environment no longer allows coverage of that because its too boring.

[1]: https://www.anthropic.com/research/economic-index-june-2026-...

The question for doomers is why is this time different? Why aren't LLM's just the tractors of the Information Age?

Saying "I became a farmer/owner/operator by getting close to the soil and using a manual soil tiller to learn about the soil. Using a tractor means there are no entry-level agriculture jobs anymore" would've been just as valid 200 years ago. So why is it different this time?

100%.

Caveat: I have kids.

But isn’t this also just bad implementations of AI? “It got a stupid simple answer wrong”, seems like a gpt-4o on a buggy RAG implementation from 2 years ago problem.

Hard to fathom a modern model would get any of this basic stuff wrong. I know accountants using it for (a lot more complex but similar problem) tax law parsing and it going through flawlessly.

There seems to be some bidirectional media gaslighting happening here where the doomers and glazers are talking past each other a bit.

Please give us access :) We are doing super useful things for museums and travelers and this would be a huge unlock!

Also, big fan of Dr.Li!

Wow, I have the exact same use case. Just move stuff off the floor so my vacuum robot can run and my life will improve 3-4x. Easy $2000 buy.

The matic is being sold as one with better obstacle avoidance and I have no doubt it does but the point is I want the obstacle out away and the floor cleaned underneath.

Can you speak to the tradeoffs of high ratio servers vs QDD motors? Is it power or long term life?

I can see me doing so many things with this.

>> NOT an LLM. its a small ar transforme

Super cool project! Though, aren't most modern LLM's ar transformers internally?

Super cool! How do you think you will want to sell these into robotics teams, I find so many core engineering teams in Robotics companies have a "we'll just build it ourself" attitude around tools like this.

Have you had luck cracking past that?

I would love the simplest one of these with just wheels and an arm. Could do so much with it.

Appclips in iOS are a decent shot at that. But require it to be under 15MB, which is really hard to hit, especially if you use RN, but we'll be dumping RN to make this happen as a next step.

>>The issue is that people often forget to uninstall these apps almost all the time

Fundamentally though, iOS will auto-offload the unused apps for you (you won't get any notifications). For all intents and purposes, it is an ephemeral app already.

No need to litigate this again, digital audioguides are now 20-25 year old technology and very popular.

Not persistently, the user has to constantly bring the phone up and re-enable, which as you can imagine gets really annoying really quickly.

The flip side is: I have an Audioguide app with reams of fancy of AI features. Even a basic non-AI one is we use GPS in the background to auto-trigger when you approach certain checkpoints on your tour, and we don't want you staring or constantly manually fiddling with the phone when you're on a tour in a neighborhood. This is all stuff just not possible with the web app

I get it, "I'm just traveling, why do I need to download an app which will be useless when I'm done traveling". Just delete it then, takes two steps!

Our reviews between our Web App and iOS app are like 2 sigma apart. Un-intuitively, our web app users are out fiddling with the app 2-3x more than our iOS app.

"I know you can do this with custom apps and I'm doing that but you have no clue how many non tech people beg me to share how"

Wait...I'm in tech and I'm begging you to share how. Is there a way on iOS?

I've mostly just stopped using these apps.

“the sense of superiority the experience seems to breed"

I dont know how re-reading is going to help me see this sentence differently. I am just telling you my lived experience shows me this prior is simply not true. That's all.

“the sense of superiority the experience seems to breed”

Mostly an extremely online problem, not a real world one. More so the nature of the medium in fact.

Evidenced by how you edited out the “dead” in front of VC funded when you were quoting me. You read the tone you want to read.

Yes obviously.

But just to caveat, its not YC or VC funding that is critical.

It’s the act of building something and trying to sell it that is.

N=1 but my dead VC funded YC company taught me more about the ins and outs of what it takes to run a complex successful engineering and sales operation than any number of years in business school or grinding in consultancies.

YMMV you do have to have a different less nihilistic mindset but it really is a bit of a Neo in the Matrix moment because it lets you see through a bit of the corporate veneer to see the complex mess of personal and group foibles that run our major corporations.

In the end it also ends up making you a 10x more effective worker if that’s where you end up even if just to protect your mental health by being able to explain what’s going on.

“I wonder if maybe the speed limit is too low”

Super curious how anyone could think this. It’s not a highway, and there's pedestrians crossing. The buildings aren’t dense and the setbacks are large so maybe it “feels” like you should go faster but doesn’t mean it’s safe to.

And Oakland especially has really bad driving culture so I’m glad for this automatic enforcement.

Yes we should just always give in to populism’s demands and it’ll have no negative consequences ever /s

…oops [1]

[1]: https://www.nytimes.com/2026/03/18/headway/motor-vehicles-fa...

I'm not. There are two issues in my comment.

I mean ALPR's have worked wonders in my town [1]. So I'll keep supporting them until we restructure all our cities to drastically reduce car speeds. Plus post 2020 police backlash, detective work has basically disappeared and clearing rates have cratered [2]. Police are now no longer able to (or are refusing to) solve crime "the old fashioned way", and the public has no real leverage to "make them do it".

So I get it, we're in the middle of a populist uprising against it, and it's nice to ride that wave of support. But behind the scenes, these two factors are driving adoption, and we'll keep coming back to it when the backlash eventually fizzles.

I say we should just use this backlash to put in the protections EFF suggests or fix the two underlying causes. Those are the options.

[1]: https://www.oaklandca.gov/Public-Safety-Streets/Traffic-Safe...

[2]: https://stateline.org/2025/11/24/why-clearance-rates-dont-te...

But also, I dont understand all this hand-wringing. EFF has a clear easy solution [1]: Put all the data behind a Warrant. Problem solved. No trust required.

Wanna look up where a murderer drove? Go get a warrant

Wanna look up where a robbery suspect drove to? Go get a Warrant.

Divorce case hinging on one party's whereabouts on a certain date? Warrant.

Speeding tickets are automatically filed anyways.

[1]: https://www.eff.org/deeplinks/2026/08/too-little-too-late-fl...