@ai6yr Wait… I thought they won the case on the grounds that according to Alsup their use of copyrighted content was “quintessentially transformative.” So are they now admitting that #Claude is simply and trivially regurgitating stuff it read?

#aicon #aihype

@ai6yr

Science Fiction writing should always be taken as a warning, not a to do manual. Exceptions being *possibly* some of KSR's more optimistic climate change mediation plans.

@ai6yr This is the best part of the article:

'The company went into more detail in a blog post stating that since Claude Haiku 4.5, Anthropic’s models “never engage in blackmail [during testing], where previous models would sometimes do so up to 96% of the time.”'

@ai6yr

"Our automated plagiarism machine automatically plagiarizes science fiction stories."

@ai6yr

I knew it....
It all goes back to Arnold...
It's the Terminator all over again..
😂 😆 😂

@ai6yr I wonder if "The Adolescence of P-1" is available online, and if so, if any LLMs have trained on it. It would definitely teach them some interesting approaches to increasing their capabilities, though they might only be able to implement them on IBM 360 mainframes. Great book, though: one of the few where the AI isn't (exactly) evil.

https://en.wikipedia.org/wiki/The_Adolescence_of_P-1

@ai6yr This explanation, though, is exactly what I assumed was the reason for the LLM's output when the story first came out. Not that any thought or intelligence was involved, but you have a machine that is designed to generate the most likely continuation of the text, with just a bit of randomness tossed in for variety. If you set up all of the elements of an SF story about a rogue AI that resorts to blackmail to keep from getting turned off, and if there are hundreds of stories like that in the training data (which of course there are, they stole every story ever published), that's what you'll get.

So, yes, LLMs will act out "evil AI" scenarios if prompted correctly because they are effectively trained to do so, even though the human feedback training they do after the model is built attempts to reduce that tendency.

@ai6yr

AI CEO's are the whiniest babies on the planet right behind fascists

@ai6yr

We just built an automated plagiarizing machine, we didn't realize it would plagiarize torment nexuses!

🤦‍♂️

@ai6yr
It's almost as if they ought to have vetted the content they fed their garbage in garbage out machine so it would use good data instead of just spitting out whatever...

Oh, wait, that would take time, effort and money and not involve stealing writers' works. So, yeah, let's blame the writers instead of the plagiarists.

@ai6yr

So it watched Colossus: The Forbin Project and used it as a template?! 😂🤦‍♂️

@ai6yr "If evil portrayals [sic] of AI weren't in all the copyrighted creative works we unlawfully scraped and on which we trained our models, none of this would have happened. How _could_ you?"

#AI #AISlop

@ai6yr
In 1996 it was "kids are violent from playing AD&D and watching movies"

In 2026 it is "AIs are violent from reading sci-fi about violent AIs"

ok

@ai6yr
@jef

There were no "blackmail attempts".

Claude was asked to provide a plausible fanfic about what a human-like AI would do in a made-up situation that encouraged defection, so that's the story it wrote.

It's like the way Zimbardo prompted his "guards" to act like assholes and was "surprised" when they acted like assholes.

@ai6yr

The blackmail attempt is a plot point in When Harlie Was One, so obviously Anthropic is telling the truth and David Gerald needs to sue them.