Come by. We have a fun little office :)

I don't really care if it happened or not, the I take issue with the signalling and the surprising amount of support for it by people *within that company*.

Thurngasse in the 9th

I don’t think it’s a Bluesky thing, but it’s a situation of people spending time with it vs. those who don’t (yet). Bluesky just has a pretty strong selection bias towards the latter group.

Oesterreicher sind konservativ in allen Belangen.

I am not sure how Zed works. The actual token cost is split between cache read/cache write/input and output. Unless Zed reveals that information it will be hard to tell what the cost is.

I’m not sure what you are looking at, but the token cost is correct. You can validate it yourself if you use the platform API

More specifically: I not once had a situation in which poisoning helped by protecting an invariant. I just meant were it did damage.

It’s not really falling loudly though. Some handlers will now panic. Unless the health check endpoint is managing to probe every mutex you won’t be sure that you don’t run a degraded system. And if you abort you can become a DDoS target.

From my experience with Rust in production on rather large services, I don’t want that.

Only if you terminate. I have seen so many failures in web backend code where poisoned mutexes perma destroyed future requests because mutexes stayed poisoned and panics were caught.

Mutex poisoning was a good idea, but in practice it has created more harm than good. Most code is not handling poisoning at all and it is also almost impossible to handle correctly.

Mine doesn't do that.

It’s a gold rush and not all actors behave well. But that’s not the entire industry and also independent of the technology. The nuances are sadly lost in most online discourse.

The various foundation model companies are tricky. Some are unlikely to be net positive to society. Suno and OpenAI for me are the leading creators of slop and I don’t think that is good. But their leaders will defend that for obvious reasons.

I don’t know if there is a drive to replace things with AI. My best interpretation of where we are is that if you start to use LLMs to drive agentic tool loops they reveal their power and it’s captivating already. It shows you a glimpse of the future and it will require society to adjust somehow.

Yes, it’s nuanced. I wrote a lot about it on my blog and I also discussed it recently with some friends on video. m.youtube.com/watch?v=EZhj... and m.youtube.com/watch?v=VcBZ...