https://en.wikipedia.org/wiki/Compose_key

It’s a key that makes the next 2 or more keystrokes be dead key inserts that combineinto some character otherwise impossible to type.

In my case, my keyboard had a ≣ Menu key which I never used, so I remapped it to Compose.

You’re correct, but that’s like saying along the lines of manufacturing a car is just bolting and soldering a bunch of stuff. It’s technically true to some degree, but it’s very disingenuous to make such a statement without being ironic. If you’re making these claims, you’re either incompetent or acting in bad faith.

I think there is a lot wrong with LLMs and how the public at large uses them, and even more so with how companies are developing and promoting them. But to spread misinformation and polute an already overcrowded space with junk is irresponsible at best.

I’ve been getting into the habit of also using em/en dashes on the computer through the Compose key. Very convenient for typing arrows, inequality and other math signs, etc. I don’t use it for ellipsis because they’re not visually clearer nor shorter to type.

That is not really true. Yes, there are jump instructions being executed when you run interference on a model, but they are in no way related to the model itself. There’s no translation of weights to jumps in transformers and the underlying attention mechanisms.

I suggest reading https://en.m.wikipedia.org/wiki/Transformer_(deep_learning_architecture)

I still don’t think it’s there, but development hss been fast, so a lot has changed and improved in the last couple of months.

Not exactly. Matrix 2.0 relates to the protocol (Matrix) version, which has its major number incremented due to a bunch of, well, major changes/updates to make it much better. OIDC, sliding sync and native calls are some of the new things that comprise the 2.0 update.

The server implementations are somewhat orthogonal to this. Synapse (the original Python server) is still the main implementation, and is Matrix 2.0 ready.

This. I don’t even need real “scripts”, I just plug my big drive into my laptop, start dropbear and use the shell (fish) history to get the right command. Takes about 15s to do everything :p

If you want a more continuous thing, you can look into Syncthing or something like that.

The web font would also be cached, and it wouldn’t be that big of a resource in the first place. I think being able to copy a comment’s content is more important, but whatever.

Not really, 2k is enough to have a result with a pretty low error %.

You’re totally right, my statistics is very rusty, good lord. For the ~240M eligible voters in the US, you can get roughly 2% margin of error, for the usual 95% confidence level.

My comment was a bit daft, in retrospective. Surely the polling people know what they’re doing, better than I do for sure x)
I guess it goes to show how non intuitive some statistical methods can be at first?

Isn’t 2k voters a comically small sample to draw any results from? I hope it is true, but let’s not get ahead of ourselves.

True, an RCE is always a serious thing. Just saying it’s not exactly catastrophic like others have been more so.

Yeah, exactly. Very impracticable.

musl isn’t vulnerable, as per https://fosstodon.org/@musl/112711796005712271

Mandatory link

Yeah exactly! It’s a great case of “invisible” improvements.

Indeed!

The two biggest things are lifetime extensions and inline consts. Both will allow you to write more concise code and, in the case of lifetime extensions, may help eliminate some bugs, since you won’t need to work around that limitation anymore.

I just upgraded my Lemmy instance’s hardware and finally got IPv6 support :D

We’ll be fiiiiine 🥲 starts hyperventilating