krosaen

This makes me very happy - one of the first games I played on my mac classic.

They are contemporaries - cs231n and Sebastian's first ML book came out in 2015 - both excellent!

Interesting that your experience is so different. I didn't mean strangers in my neighborhood though, just acquaintances / neighbors who I talk to sometimes - and airpods nudge the decision towards smile / wave keep walking over stop and chat.

Wearing on public transport: meh, the chances I chat with a total stranger is quite small.

Wearing on walks around my neighborhood: yes, totally see how this nudges away from spontaneous chats with neighbors / acquaintances as we pass each other and wave, but don't stop perhaps due to the friction of removing airpods and sense we may be interrupting each other.

Sometimes if I see a neighbor I know up ahead, I will preemptively remove my airpods to open the possibility of a chat. Most of the time just a hello, but sometimes a nice catch up. With the airpods, very unlikely to chat.

The CEO of Toyota Research Institute, Gill Pratt, had some good (albeit self-serving) points about the benefits of hybrids and PEVs to complement EVS a few years back that I bookmarked and keep coming back to:

https://web.archive.org/web/20210727090309/https://medium.co...

https://web.archive.org/web/20210825054702/https://medium.co...

> Maximizing the benefit of every battery cell produced requires that we distribute them smartly.

> This means putting them into a greater number of “right sized” electrified vehicles, including HEVs and PHEVs, instead of placing them all into a fewer number of long-range BEVs, like my model X. This is particularly important because presently it is difficult to recycle the kinds of batteries used in BEVs. If we are to achieve carbon neutrality, we must pay attention to all parts of the “3R” process — Reduce, Reuse, and Recycle.

> For example, we hardly ever put gas into our RAV4 Prime PHEV, which has a battery ⅙ as large as our Model X BEV. For the same investment in batteries as our single Model X, five other RAV4 Prime customers could reduce their carbon footprint too.

> But research is showing that social media and smartphones have made us addicted to screens from a young age. It’s taken a toll on how much time we spend together

Agree completely, but it has also lead to widely held pessimistic beliefs like

> But how do we find meaning when the climate is warming, politics is broken, and technology serves profit over people? We can’t think about thriving; we're merely surviving.

I anticipate downvotes, but I seriously suggest reading https://en.wikipedia.org/wiki/Enlightenment_Now for some perspective about long term positive trends, and insights into why we over weigh short term negative news to the detriment of our mental health.

Ha whoops. But yes, the ads on the neal site I meant to link to had the aforementioned problem

Assuming you go down the path of allowing online anything, seems like, after doing your best with parental controls, the most effective thing is time boxing screen usage. Only so much can happen in, say, 2-3 30 minute sessions throughout the day, and the chances of a kid deciding to blow their precious minutes responding to some random person seems much lower than if bored and checking messages idly. Being nearby during a healthy sample of sessions to have a pulse on what's going on helps too - usually pretty obvious what they are doing.

But I share the frustration of the author with how unreliable the controls are. Apple screen time controls routinely stop working - especially the one that only allows access to a finite list of websites. I need to check the browser history every week or so to confirm it is still working, and do some dance where I turn off controls, reboot, then turn back on every once in a while. The reason this particular control is important to me is that, even starting with something as pure as neil.fun, ads on that site have proven to be a few clicks away from semi-pornographic sites - it's terrible! And yet, turning off all internet access is such a coarse decision that limits access to things that are generally informational / fun / good (like neil.fun, or sports facts sites).

Anyone have experience with https://valetudo.cloud?

Seems like could be a good solution to using best rated (chinese) vacuum's while mitigating privacy concerns.

I am bummed that US robovacs aren't that competitive. Rooting for Matic, though currently don't seem to be as good as Dreame / roborocks [1] (can't go under furniture, apparently take longer to clean same area, tout "vision only" as a feature while charging more - you would think having fewer sensors / no lidar would bring costs down).

[1] https://vacuumwars.com/matic-robot-vacuum-review/

Interesting, yeah, my instinct is that the sensor weight delta would be negligible but perhaps not.

You probably don't want a long range EV then?

Saw this too on mobile - I think it's an ad - I requested a paid ad-free version elsewhere in the comments

Neal.fun is good clean fun - my kids love it too. Neal, if you are listening, would pay for an ad-free version (I already bought you some coffees too).

The founder of Zingerman's (famous deli and family of businesses in Ann Arbor) description of servant leadership is a bit more complete and overlaps heavily with what the author of this post is advocating for:

https://www.zingtrain.com/article/servant-leadership/

If there is a primitive not currently supported (say running a temporal workflow service) is it possible to define a new primitive for this? Just wondering what it looks like if/when you need something not currently supported.

No so much about why collaboration sucks as and argument for how important direct ownership / responsibility is. Good post.

This reminds me of the pains taken to reproduce original Game Boy Advance fidelity in the chromatic by mod retro https://modretro.com/blogs/blog/display-the-hard-way

I like the idea of the `defer `keyword - you can have automatic cleanup at the end of the scope but you have to make it obvious you are doing so, no hidden execution of anything (unlike c++ destructors).

Hammock driven development :)

https://www.youtube.com/watch?v=f84n5oFoZBc

This looks pretty awesome, and excited it's not only a cloudflare product (Cap'n Web exists alongside cloudflare Workers). Reading this section [1], can you say more about:

> as of this writing, the feature set is not exactly the same between the two. We aim to fix this over time, by adding missing features to both sides until they match.

do you think once the two reach parity, that that parity will remain, or more likely that Cap'n Web will trail cloudflare workers, and if so, by what length of time?

[1] https://github.com/cloudflare/capnweb/tree/main?tab=readme-o...

We've lost our sense making ability in our coverage of AV progress. Setting aside Elon's polarizing behavior (which is a big thing to set aside) all of these things seem true:

- Elon bullshits wildly and sometimes delivers

- FSD is far ahead of other available personally owned autonomy - no other system even attempts to drive on surface level streets. FSD works better than you would think - I've been in several flawless rides across town in conditions I didn't expect it to handle.

- FSD doesn't work well enough to rely on it yet, so what's the point for real consumers who don't want to sit there nervously hovering their hands over the wheel even if it is impressive and you might have to ride several hours before anything goes awry.

- We can't really know how close FSD is to being reliable enough because all we have is marketing claims from Tesla, fan boy clips on YouTube, and haters who can't seem to discern where FSD really is ahead, even as it falls far short of hype

What I wish we had was a common metric audited by a third party reliably published in terms of hours until disengagement or something like that across all systems in various conditions.

I find this post interesting independent of the question of whether leetcode problems are a good tool for interviews. It's: here are some kinds or problems constraint solvers are useful for. I can imagine a similar post about non-linear least squared solvers like ceres.

Reminds me of https://github.com/bhauman/devcards

Yeah probably, if I didn’t get it bundled I may very well be using spotify. But Yt music doesn’t suck and its YT integration is mildly nice.

I pay for YouTube premium (which gives me YouTube music too) and am happy with the lack of ads, even though many creators still mix paid sponsors into their videos. It seems the creators are motivated to keep things minimal or they will lose engagement.

What I am not happy with is a lack of control over the homepage and recommendations. I would really like to be able to easily block channels from ever showing up, but you can only sort of do this if you click "don't recommend this channel anymore" from the homepage. But you can't do this if a video shows up recommended from another video. And overall, it just feels like they are spending so much effort trying to get me to watch the next video instead of enjoy the one I am trying to watch.

For my kids, I came up with an ad hoc policy where they can watch from the homepage / recs on weekends but during the week have to stick to a personal playlist they can only add videos to on the weekends. This removes the algorithmically driven addictive nature of YouTube and unsurprisingly they end up moderating their use of Youtube within their alotted screen time much better. It distinguishes between, "I want to watch this" and "I want to pull the slot machine lever." But I would be a lot happier if I could better curate access to content for my kids too. Youtube Kids sucks, it ends up filtering out a bunch of interesting stuff like carpentry and nature content that hasn't been marked "for kids" in favor of videos of kids shopping for toys and stuff.

ah whoops, thanks.

but did you see that now Wang is leaving as part of this?

Anyone know what scale does these days beyond labeling tools that would make them this interesting to Meta? Data labeling tools seem more of a traditional software application and not much to do with AI models themselves that would be somewhat easily replicated, but guessing my impression is out of date. Also now apparently their CEO is leaving [1], so the idea that they were super impressed with him doesn't seem to be the explanation.

[1] https://techcrunch.com/2025/06/13/scale-ai-confirms-signific...

After researching a few apps, I have found Piano Marvel to be pretty good for automating practice of both scales, arpeggios and ear training as well as having a bank of songs. I use the web version, and the interface is clunky, but it’s the closest I have found to a math academy like took for piano.