Not sure how many of you use ChatGTP, but you might want to scale back.

[redcrew@mstdn.social]

@stefan Never used it, and have no future plans to use it.

[natanox@chaos.social]

@gilraen79 @jonas175 @stefan Same, however you of course still have to be cautious how and for what you're using it. It's still an LLM that will slop the slop. And it's crucial to immediately disable the picture generation feature to not accidentally have it create pixelized nonsense.

If the bubble pops Mistral also has a higher chance of surviving given their focus on on-premise and very compact models for "edge" devices. They've realized the current tech has pretty much reached a plateau.

[pcat@mstdn.plus]

@stefan Then chuck it. If enough people turn their back on that piece of shit Musk, he'll wither away. Treat X and the Grok shitosphere like Tesla. Cybertruck sales are down 50% from this time last year and people are rolling their eyes when that human nutsack says anything.

[michelletaylor541@mastodon.social]

@stefan I already stopped using it. Found Claude better. But now I only use Le Chat

[heri@net.miaumuh.ch]

@jonas175 @stefan

Mistral is good, indeed. I like also Deepseek, claude.ai and public.ai from ETH Zürich.

[gilraen79@mastodon.social]

@Natanox thank you for the advice! I'll check it's not enabled immediately

[stefan@stefanbohacek.online]

FYI, not a ChatGPT user myself, so no need to reply with alternatives people should be using instead. Thanks!

[holdenweb@freeradical.zone]

@stefan @BernieDoesIt Slop trained on slop.

[staringatclouds@mstdn.social]

@ShaulaEvans @stefan AI centipede maybe

[olicorne@mastodon.pirateparty.be]

@stefan
IMHO given that those are quotes by LLM, the true issues is with the search engines (or grok's terrific SEO) and not the training process of the models.

[wesley@defcon.social]

@stefan a confidence-building response. "When asked for comment, a spokesperson for xAI, the owner of Grokipedia, said: 'Legacy media lies.'"

[gilgamesch@mastodon.social]

@stefan Habsburg AI incoming.
It's finally happening.

[madsenandersc@social.vivaldi.net]

@stefan here is the thing as I see it. People has to stop seeing AI as an oracle and start seeing it as a slightly more useful search engine.

I have a local instance of gpt-oss (20B) running on ollama, querying it through Open WebUI and doing web searches through a local SearXNG instance.

It's absolutely fine for tuning a job application for a specific job posting. It works really well as a translator of texts from Danish to English.

Although it warns me that it is no medical expert, it can even help sorting out symptoms without falling into Doctor Googles default diagnosis trap ("It could be deadly but if you are alive in the morning, you should see a doctor").

Seriously, people need to get their head out of their behind and start treating this as what it is: A computer program that can help you with tedious jobs and help find answers to even somewhat difficult questions, but nothing more. It is really good at helping you with something that fits a pre-defined pattern, but it sucks at being creative.

Oh, and yeah - don't use a public version outside of the EU, where it is under strict GDPR laws, unless you like to live dangerously. Mistral for the win.

Seriously, the sky is NOT falling.

[mls14@social.vivaldi.net]

@stefan It destroys the planet. It makes you dumber. It feeds your private info to tech billionaires. Nobody needs this, and anybody using it for any reason should stop immediately.

[webjo@mastodon.social]

@jonas175 @stefan

I trust only CatGPT. Hard facts, often painful.

[aurelia@social.treehouse.systems]

@stefan this also goes for the very ethical and constitutional claude, by the way

[dexternemrod@troet.cafe]

@webjo

@jonas175 @stefan

Well "le chat" is french for "the cat" soooo ...

[dexternemrod@troet.cafe]

@stefan
Did I read this correct: A llm got poisened by another llm?

[aburgerabz@mastodon.social]

@stefan what a surprize!

[stefan@stefanbohacek.online]

@dexternemrod Yep, basically.