nasretdinov

I wonder if you have a lookup table for values that _rarely_ change you could use the same approach (with maybe one caveat that you'd need to track the lifetime of the old variant of the hash table)

Not all environment variables will be passed through (depends on the server configuration), but stuff like TERM should

> It's totally fine for basic UNIX tooling, but pretty painful if you're trying to make nice interactive TUIs.

This sounds like a deliberate choice from Apple TBH because they, well, like their UIs and don't want to promote TUIs necessarily :)

It's funny that it takes checks notes a video game card manufacturer to make a good sandbox for AI agents and not the AI companies themselves...

Well I'm obviously not comparing to average human intelligence (and negligence), but that's actually a good reminder about one other important aspect of how living organisms are very different from machines -- humans have very high range of intelligence, strength, etc, and the best humans are usually much better than the average in all areas of life. Computers and software are much less diverse.

Unless something dramatic changes and it'll become cheap to "constantly re-educate" the models to learn the new fundamental facts about the system (ideally with a small lag of maybe a few days like with humans) I see no reason to think that LLMs can have a profound understanding of how business systems work of a remotely comparable quality to humans. Just adding the context about the system already reduces it's capacity significantly according to what I can tell. Not having large context helps somewhat at a cost of having to explain lots of task-relevant things to LLM over and over again...

If you don't need SSL you can't use HTTP/2 or HTTP/3. It doesn't make sense for externally facing services but sometimes is ok or even unavoidable for local networks

It sucks that this changes the behaviour under the hood that you previously coukd rely on the fact that closing the response body lets you skip downloading the HTTP body entirely, but I do like the tradeoff here, since it also limits the download time to just 50ms you'd not get overwhelmed if your network can't keep up with all the cancelled HTTP bodies, you'll just wait a little bit of traffic instead.

Honestly the network-attached storage only ever made sense with HDDs, it's ridiculous to have to pay 10x penalty for both iops and latency just for some "operational flexibility". You get way simpler and much more robust architecture if you just stick to local NVMe and replication -- not that you can forgo one in the (also replicated) network block storage solution anyway...

Copying GC will make swap even less efficient, no?

Unfortunately that's somewhat expected - regardless of implementation, the sole fact that the GC needs to somehow walk the entire tree to trace still referenced sections makes swap highly impractical -- even if you had not had 40ms stop-the-world pauses the LRU cache used by swap would get thrown away at every GC.

I believe that's actually the same reason why Apple stopped using GC in their frameworks in favour of automatic reference counting.

Well, you don't have to use the proprietary third-party SDKs in your apps... If you want analytics you can either implement your own, or use some open-source ones that aren't dependent on some third party backends switching some features on for everyone all at once...

If only there was any way to avoid this, right..? (I don't mean from Google's side, that as well, but that's not the point)

I must say I find my colleagues who use AI instead of answering the damn question equally frustrating, and unfortunately they're in another country so I can't throw them into the ocean

The risk with any cloud is that while your data might be physically intact, you can still lose access to it for a variety of reasons (e.g. being born in a wrong country)