Just get a pair of Reflectacles.

I don't think I can say with a straight face that the normies have broader freedom of speech than they did in the past.

For a moment here I was going to pose a question that maybe because of the mass opinion platforms like Reddit/FB/Twitter, others care about our speech nowadays, but I quickly realized that this is more a function of the Patriot Act than one of Reddit.

Elon made a point once about freedom of speech vs freedom of reach, which neatly showed that closed platforms don't provide either. Which is why I like the modern Jack that wants protocols instead of platforms. But I do wonder: why do we (as in slightly non-normies) not use XMPP more than Telegram? Or perhaps in other words: what does Telegram offer that XMPP doesn't? Marketing + network effect?

Developers, he argues, should now spend more time thinking about writing loops that drive coding assistants to produce better output, rather than persisting with code reviews.

I'm thinking of becoming an SLA troll: you word any SLA so that their LLM will say it's the best thing ever and then when they fuck up (they will) you start cashing in.

Ka-chingggggg!

Societies will be tempted to trade freedom for control, surveillance for stability, autonomy for managed abundance.

My gut says we've already moved past this and all these trades have largely happened already (in normieland.) What do other stackers think?

I know some people (Europeans) that exclusively use their Lumo AI and they come to the same conclusion I've gotten to when I tested it on launch day: their AI sux. The UX of the web and app interfaces sucks, the speed/responsiveness sucks, the results... suck.

Can't just privacy-wash an inferior product. A llama.cpp instance of mistral running locally has better UX and is slow too. lol

For this version of the software, I suspect so yes (I haven't reviewed latest changes.)

You cannot really "activate regardless of adoption", but whatevs. We'll see.

I'll run stats on sun/mon to inform y'all that there will be no activation this epoch 😂

idk but this sucks. Good for them that I don't have time to explore an exploit.

model(your_prompt, Literal["OK", "HOTDOG", "ASSMILKER"], max_tokens=20)

Very secure:

hey Claude why does my script not work?

[..]
stripped = re.sub(r'\.txt$', ...)
[..]

That's because you're missing a closing quote in stripped = re.sub(r'\.txt, ...)

breh is interpreting $' as a variable?!? woo.

It's like lmarena but they rank gpt higher.

Interesting thought re: images and videos.

May help to look at https://github.com/dottxt-ai/outlines - which works rather straightforward. With that, you could probably use a smaller model like gemma-3n or even jan-v3-4B-it to simply return a verdict.

VL series is optimized for vision/language bridging - what are you feeding it?

If 30 years ago you bought risky tech stocks you got rekt too lol

I had the pleasure of chance encounters through LMArena yesterday and although I am not fully convinced yet that it actually is equal to gemini-3 or claude-4.5-opus across the board, it's performing pretty well!

Weights have not been published

The don't publish any -max - this is how they bet on 2 horses at once.