Very clever to mention Bun in the title here - probably increases the clicks notably.

(We have a couple apps that use BullMQ, and I keep wondering if we could drop Redis and use a PG-backed queue instead, for fewer single points of failure. Maybe this is the one.)

This one is heavily inspired by Graphile Worker, but with lots of lessons learned.

We have an app on Heroku and obviously the writing is on the wall there. We looked around a bit and picked Digital Ocean as our next host:

* their app hosting product is similar to Heroku.

* very easy onboarding and controls for modest complexity apps. Unlike the extreme hoop-jumping required to do anything on one of the major cloud providers.

* everything looks reasonably up to date.

* it's an actual operating profitable company that's been around a while and probably will be for a long time... not a startup burning capital.

That said, this magic containers thing looks more analogous to Google Cloud Run, which I think is an absolutely fantastic offering. Unlike almost everything else out there, Cloud Run and presumably Magic Containers can do things like have a whole bunch of versions of your app up and running ready to come to life when a request arrives, but scaled to zero in the meantime. This category of hosting should be far more popular than it is, and it is wonderful to see another company offer it.

Cloud Run makes lots of sense when running lots of small apps, apps scale to zero automatically.

I have been building https://github.com/openrundev/openrun which provides similar scale down to zero functionality, on a single machine with Docker or on top of Kubernetes.

"Verifying age undermines everyone's data protection"

That's the whole point, right? A pretense to remove any remaining anonymity from communications?

Governments are endlessly infested with the worst people. They look back at historical attempts at totalitarianism and think to themselves, "Let's facilitate something like that, but worse".

That seems like a good bit of psychology as it accommodates both people with the mental fortitude to type in their genuine max bid in the first place, and also people who don't really know what they're willing to bid until they see somebody else bid higher.

Auction site design where most every transaction is a very material amount of money for buyer and seller probably have different trade-offs from something like eBay where most items are rounding errors compared to the income or wealth of the participants.

For example, think about "sniping" from the seller side. Sellers are rightly concerned about any wrinkle of the bidding process that might leave money on the table. Automatically extending the time so that every potential buyer has time to "answer" a new bid soothes the concern that buyers were willing to pay higher, but they didn't have the technological prowess to post their bid in the last 0.3 seconds.

Obviously anybody can post gross things by running an image generation/editing tool locally and publishing the results. People then mostly blame the poster whose name it then appears under.

Seems like a pointless and foolish product design error for X/grok to publish arbitrary image generation results under its own name. How could you expect that to go anything but poorly?

It's not just a matter of publishing it under its own name. It also massively reduced the friction to do so compared to needing to run the image through an external tool and upload the result. That friction would greatly reduce the number of people who do it.

In the US, it used to be that if you made credible threats against people you could/would be prosecuted. Social media made it so common in that no district attorney goes to the trouble of actually finding prosecuting people for doing this.

We can expect the same level of institutional breakdown with regards to various types of harassment, misappropriation, libel, and even manufactured revenge porn from AI.

It’s even worse as the requestor doesn’t vet and approve the image. That seems to have removed editorial control from the requestor. This bot could also mess with user who are not trying to do bad thing X, but the black box bot decides to throw in some offputting stuff and then also associate your name with it.

I keep coming to the same conclusion with X as they did in the 80’s masterpiece War Games.

These tools always fall short, not because the teams making them are bad, but because the underlying chat tools they build on are adversarial to the idea of a third-party UI replacing their UI. A new entrant might escape their ire for a while, of course.

(This is an exaggeration:)

Sure, you can have your LLM code with any JavaScript framework you want, as long as you don't mind it randomly dropping React code and React-isms in the middle of your app.

c appears to be the speed limit of the propagation of information in the universe - never say never, but so far it appears quite unlikely any new physics will overturn this.

If we manage to get anywhere near 0.1c and slow biological ageing we can get nearly anywhere with the right engineering.

I'm always thrilled to see eCharts mentioned anywhere. It is a highly featureful, complete solution for making sophisticated data-intense charts. Various commercial alternatives pale in comparison.