@michal #JustSayNo

@michal

Re. No theft: They must still be trained on vast quantities of text, in order to 'understand' the input, even if they output JSON rather than human language. I doubt their training corpus is much more ethical than those used for LLMs?

Re. No hallucinations: They could still hallucinate things or be influenced in weird ways from things in the input, affecting the decision output. And then you're left with just black box decisions, without any record of why they were reached.

@michal I forget who said it, but "AI companies are spending billions to make millions." And they want more investment.

This will end badly. Just like Tulip Mania, the South Seas Bubble, and all delusions throughout history.

@michal When you lose 42 billion and still Reuters writes about your "sweeping vision", that's quite a success in my book.

@michal yeah I don't hate it, and I usually hate change

@michal Why would it be dumb? Maybe not cost effective.

@michal exactly. They could've blocked all those ads for scams, pyramid schemes, fake AI politicians endorsing various grifts. They just didn't want to, I guess, for some reason.

@b0rk whatever magit does 😅

@b0rk

usually the `git push -u` because that's what git instructs me *right there* in the error message when I try to push a branch that lacks an upstream.

However, if I *remember* and know ahead of time, I'll edit my .git/config and manually create the branch/remote/upstream linkage just to avoid having git yell at me (and because I never remember the `push -u` syntax until it does yell at me, but I can easily see the pattern in the config file)

@michal no thanks I hate it, but you do you

@michal

You don't want to participate if you have to ask

@announcements Question: Can I use this space to ask about developing/deploying/operating LLM and alike locally? Or in a self-hosted way?

Another question: does this include other technologies adjacent to it, like machine-learning? Like Discussing OpenCV or running machine-translations with Weblate, etc.

I understood that we shouldn't use this space to promote monthly subscriptions to a proprietary black-box in someone else's server. But can we talk here about the black-box if it's (semi)open and running locally?

@announcements

why am I the only one who thinks it odd to oppose ai cause " controlled by a very small number of extremely large companies "

and at the same time not be opposed to using computer chips made by Intel/TSCM/AMD ?

how on earth are the two techs different ?

@announcements

The first rule of BSD Cafe is: we do NOT talk about AI.

@stfn

Mydło weź. I powidło.

@michal Będziesz mógł sobie 24h odebrać awizo.

@michal to już zrobiliby odwrotnie... odbiór awiz, nadawanie listów itp na każdej stacji benzynowej orlenu... brzmi lepiej ;-)

@michal

read aloud.

chicks dig it.

@aga

How did it handle it? :)

@michal I can recommend Taxi, if you haven't seen it.

@michal Everybody wants surveillance, even the EU... I think we're putting them on a too high pedestal. See ChatControl, see age verification etc.

@michal The whole thing was a metaphor, invented by Intergalactic Digital Research, to mimic how people worked in the 70's. It really makes no sense now. If it ever did.

Can't watch at work, but I'll run this when I get home.

@michal Oh, my. That's my favorite cola brand, and it's almost impossible to find now. Plus I mostly don't drink sodas now (too fat, and sugar won't help). But if I could get some to keep in the fridge for special occasions, I would.

(I'm also finding it hard to get Little Debbie snack cakes. I love the Swiss Cake Rolls, though I'm not sure what the Swiss have to do with it. ;) )

@michal

The RC abbreviation makes for good Unix fit: /etc/rc.cola is a perfectly reasonable file

@Moosader I dunno, I find great amusement in the video I'm currently watching, describing it like a dysfunctional office :v

@tootbrute John Abercrombie, Dave Holland, Jack De Johnette : Gateway and Gateway 2

@stfn

Stado to stado, zawsze razem 💪

@mos_8502 what happened to your current instance?

@mos_8502 bitbang is pretty much dead btw, running a 2 year old version of mastodon (without even security updates) and I don’t think it’s accepting signups

@thomasadam get out more to get more cow mugs and crocheting stufff? 😂 (cool cushions on the seats btw :)

@thomasadam
Is that a proper stereo I see in the background? Which reel-to-reel tape recorder is that? You are either dating yourself or you are a millennial with great retro aspirations!

@bert_hubert.. It could, if its owners would provide it to researchers doing data-heavy pattern recognition research. However, this will not happen, considering these owners provide server capacity to the highest bidder instead of the best cause.

@bert_hubert it is well on the way to *diagnosing* most diseases, anyway. https://mastodon.social/@hyc/117036731217520564

Howard Chu @ Symas » 🌐 2026-08-04
@hyc@mastodon.social

Rethinking EEG-Based Disease Diagnosis: Decoupling Instance Representation Learning from Subject-Level Supervision

EEG-based disease diagnosis requires one prediction per subject, yet common pipelines segment recordings into short instances...

We propose BridgeMIL (Multiple Instance Learning), a two-stage framework that decouples instance representation learning from subject-level supervision

https://arxiv.org/abs/2607.27274v2

Built on #LMDB