How does it save tokens?

I've heard lots of people use it to actually find out what you were being banned for even when the company absolutely does not want to disclose it.

We should create gdprdelights.eu and list of these use cases.

Or maybe it's just lazy programmers, wouldn't be the first time.

I'd like to see them try to increase prices as long as open source models that are roughly in the same league are available

I have yet to find a messenger with a better UX than Telegram. And I have tried lots in countless attempts to get away from this badly encrypted russian App.

I agree, I personally really dislike discord - I do get that it works better for certain group sizes, but still.

Given that most of the clients are open-source, I wonder why signal or someone else didn't just change the "backend"

I really don't get the appeal. I mainly use Signal, though I also use Telegram a lot, and Telegram seems a lot more dodgy. I constantly get spam on Telegram even though I never get spam on Signal.

Well, I'm fairly sure that Telegram has substantially more users, so not getting spam is more likely due to that only.

Theoretically yes, but in reality I have found that often it's much easier to just fix something locally or with a workaround rather than jumping through all hoops required to get somebody in another team or even another company to understand and agree with you on your (or your teams) issues, let alone fix them in a satisfactory way.

When I downloaded it via F-Droid earlier today it did what wanted. Unfortunately Android Auto is only enabled in the Play Store Version, so I downloaded that and there it went the same as for you. At least it immediately prompted me to download my local tile set when I tapped the geolocation button to zoom in.

I still feel that during agentic workload sometimes it would be nice to have the model identify it is veering off the main track, send out a "keep the cached states and tell me which they are" command to the inference server, do the side thing (such as handling an error that plopped up that has not that much to do with the main task) and return back to the cached state with just a comment tacked at the end to say "oh and btw I fixed DNS" instead of having the DNS debugging inside the context window now. Maybe other harnesses just steer the models more towards using subagents for such tasks and my pi is misconfigured. I can use the tree feature, but having insight into what's cached would be nice there.

How's the battery holding up during vibe coding sessions or occasional LLM usage? I've been thinking about getting a MacBook or a laptop with a similar Ryzen chip specifically for that reason.

Currently I don't use vibe coding or even code assistants, so I can't speak to how the battery fares when doing that sort of thing. I don't know how much or how intensively they need to run the underlying LLMs.

For chatting with LLMs via ollama, I've seen total power usage go to about 50W (on an M3 Max) while the LLM is active, which is about 3x-4x power usage compared to just idling with browsers and editors open.

So I'd estimate about 2-3 hours of continuous LLM use on battery. Because I have enough RAM spare, at least there's no need to keep shutting down and reloading models.

I haven't really pushed it to find out how long they run on battery, as I haven't used LLMs all that much.

I'm more interested in the underlying operations of how they work, investigating novel model architectures and techniques, and optimising performance, than actually using them as an end user :-) Similar to how I enjoyed writing game engines more than playing games :-) Maybe I'll get into using them more in future.

The paper does mention that you can have your ssh keys signed by a ca, so in a company the it staff could configure everybodys os to only trust ssh keys signed by the organization.

> you can have your ssh keys signed by a ca

Good idea. That way when your CA private key leaks (the key which we never ever rotate, of course) the bad guys can compromise the whole fleet and not just one server. Bonus points if the same CA is also used for authenticating users.

>That way when your CA private key leaks (the key which we never ever rotate, of course)

As with X.509, any serious usage will involve a hardware security module, so that compromise of the CA host does not allow the key to be leaked. You'd still have a very bad day, but it can be mitigated.

I do think it's a fairly significant flaw that SSH CA doesn't support intermediate CA's (or at least didn't last time I looked into it) to enable an offline root CA.

>Bonus points if the same CA is also used for authenticating users.

The SSH CA mechanism can be used for both Host and User auth, yes.

Keeping in mind, in a real use case this would be tied to something like active directory / LDAP, so you can automate issuance of ssh keys to users and hosts.

Systems configured to trust the SSH CA can trust that the user logging in is who they say they are because the principal has already been authenticated and vouched for by the identity provider, no more manually managing known_hosts and authorized_keys, or having to deal with Trust On First Use or host key changed errors.

You can also set the CA's endorsement of the issued keys to fairly short lifetimes, so you can simplify your keymat lifecycle management a great deal - no worrying about old keys lying around forever if the CA only issues them as valid for an hour / day / etc. .

Overall I think you still come out ahead on security.