TobTobXX

Well ok, not dead. Dead sounds final and I still think that we will get more IPv6 deployed in the future. And I set up all my servers as IPv6-first. So no, I don't think IPv6 is dead.

But I'd claim it failed completely. Many very public services (eg. github.com, sentry.io, office365.com, archive.org) outright don't support it. And the fact that infrastructure providers (like mobile network carriers) don't support it either, proves that it is far from "alive". In fact, I'd say it is deader than MMS, Fax and landline. For these I can still get service. For IPv6 (on the phone) I can't.

/me searching through all the mobile network providers in Switzerland (a technologically advanced country) and not finding any.

Seriously, this drives me mad. There is NO WAY to get IPv6 on mobile networks in Switzerland. Neither Swisscom, nor Sunrise, nor Salt, and of course no reseller.

And Android handles this particularly poorly. I have some servers set up in my tailscale network on private IPv6 addresses, but android doesn't query AAAA addresses if it has no ::/0 route, so I HAVE to route all my traffic through my VPN which often creates other complications (with corporate wifi etc.)

> Gemini Carbon

> [Header image is a chunk of ice]

???

Is LFM3-Omni-2B worse or roughly equal at vision tasks compared to LFM3-VL-2B?

In other words: Should I select the model that matches my desired capability set or just use Omni?

I think that this is Kimi K3.1 Pro on max too. Unfortunately, AA doesn't seem to test for other effort levels for models not from Anthropic/OpenAI.

> taking tool calls, code modifications, etc, and dumping them in favor of a summary

Doesn't that destroy the cache? I find that caching significantly sped up my Qwen, especially on said larger contexts.

A skill should only document behaviour the LLM didn't/couldn't exhibit on its own.

So you take your failed case (eg. working with gdb or whatever), write a skill and then test for that failed case.

Can you tell more about your experience with Ornith? I've come across it, and the benchmarks on its landing page are unbeleavably good. But it isn't featured on more established benchmarks like ArtificialAnalysis and it's not on OpenRouter, so I wrote it off as scam.

No, I think both, political and apolitical decisions can exist.

If you ban a project because its CI pipeline mines crypto and that ia a drain on your server, it is an apolitical argument.

If you ban a project because it drains recources because an AI agent is commutting multiple times a second, that is apolitical.

If you ban a project, because it hosts content that doesn't align with your ideals since it uses cryptocurrencies, that is clearly political.

Sure, there may be a small grey zone. But there are swaths of decicions that are clearly political/apolitical.

That tracks. As a teacher I sometimes find myself conversing with multiple kids simultaniously as well. If it's nothing too deep that requires full focus, it works. (Though I do find it tiring and avoid it.)

Too assymetric IMO. I have no problems winning with hounds even on hard, but I have no chance of winning except against easy.

Only side effect I ran into is that journalctl couldn't find the logs of a previous boot.

I don't work in tech (school teacher), so the main way I interact with tech people is online.

IME, everyone I meet offline has some low-level caution about AI taking their job, but uses AI and is amazed by their capabilities and is glad for this tool.

Most ppl I meet online are strong anti-AI advocates.