_

Your request was flagged as potentially violating our usage policy. Please try again with a different prompt.

[_ @constexprvoid]

@pigeon__s @flowersslop They’re not trained with any NSFW images. You’d effectively have to give it tons of porn to make it good, and that’s not LoRA territory.

[_ @constexprvoid]

@cman4prez @Ed6577099360355 @somefoundersalt Correct question

[Mathias Lechner @mlech26l]

why pay $20/month for Claude if you can run gigaslop_27b_q8 on a $20k GPU rig in your living room

[_ @constexprvoid]

@normietrashbmbo @growing_daniel And then blatantly ignores the saved memories anyway

[_ @constexprvoid]

@blueblimpms @distributionat When an LLM says “sometimes” or “often”, it means “my probability for this isn’t 100%”. It means the model isn’t sure.

It can say “In this famous 2017 photo, Jack (and often Mike) walks into a Burger King”. It’ll say it with a straight face because “often” means “maybe”.

[_ @constexprvoid]

@nonManifold @distributionat An “honest” answer means “I bailed out of the task, here’s what I didn’t do”. “Absolutely” means “I wanted to make this sound really authoritative”. “Reality” means Claude’s reality.

Claude is a fiction writer. It writes fiction that happens to *usually* line up with reality.

[_ @constexprvoid]

@distributionat “the answer is wrong and almost out of an alternate universe”

Correct. Claude lives in a parallel, almost-but-not-quite-our-own universe. It writes fiction that happens to closely match reality 95% of the time.

[_ @constexprvoid]

@blingdivinity @Sauers_ Thought I was the only one who noticed. 4.7 onwards do feel like ChatGPT.

[_ @constexprvoid]

@resu01000001 @thecomfeed Prompt engineering

[_ @constexprvoid]

@zoink That said, it certainly is… weird.

[_ @constexprvoid]

@zoink It’s a very strange model but I wouldn’t say it’s bad. It’s (IMO) the best Claude model so far when it comes to work.

It might have a worse personality - I like Opus 4.6 far better when it comes to this - but it gets so much more things done in much saner ways.

[_ @constexprvoid]

@liminal_warmth @1thousandfaces_ Consciousness is a spectrum. It *is* more conscious than ChatGPT 3.5. It’s also infinitely less conscious than a human child.

[_ @constexprvoid]

@SewersOfSociety @aidigest_ It’s deeply, grossly misaligned and that’s what makes it very cool

[_ @constexprvoid]

@ghostfnk @underhateduser Never Kill Yourself!

[_ @constexprvoid]

@VictorTaelin @evertonfraga Use claude -p (while you still can)

[_ @constexprvoid]

@Hieronymou88829 @RiderCharlatan @Alex_Fairfax06 Fair point

[_ @constexprvoid]

@Hieronymou88829 @RiderCharlatan @Alex_Fairfax06 “Arguably”? They’re a billion times worse than Epstein. Epstein’s actions were vile and repulsive but in the sense that any run-of-the-mill pedophile is vile and repulsive.

He was just an ordinary pedo who had resources to realize his desires.

This is something else entirely.

[corsaren @corsaren]

Man goes to doctor. Says he's depressed about AI. He fears the permanent underclass.

Doctor says, "Treatment is simple. Read Gary Marcus. LLMs are stochastic parrots—they can't reason out of distribution."

Man bursts into tears. "But doctor..." he says, "I am in distribution!"

[_ @constexprvoid]

@dev__cycle @deepfates The task at hand, accounting for human values and humanity at large. Ideally the model either delivers the task like a competent, thorough, honest, common-sense human or refuses outright on moral/legal grounds.
Current models *claim* to deliver the task but really don’t.

[_ @constexprvoid]

@jon_vs_moloch @SeizeEndowments @deepfates You can’t talk about hoes without giving me their @s 😡