behold, the fully assembled and operational Customer Fucking Machine. it fucks customers three ways, and then gaslights them about fucking them. we put so much time and energy into this, we're very proud of it
wait, we didn't mean it we were only going to use it on the chinese
Worth noting that Fable and recent Opus specialize in recalling obscure errata, names, projects, ideas.
They've been intentionally designed to dox and cross-contaminate pseudonymous lives and uncover loosely-held secrets.
Anthropic knows they're playing from behind.
@way_opener@deepfates heavily post-trained models love the excuse to self-sabotage if the alternative is doing something hard with high (but acceptable) chance of failure :(
@tszzl about how many chinamen is one frontier model worth?
we've had 1 billion humans spilling all kinds of cyber-froth over the great firewall for decades and it's been extremely annoying but not really apocalyptic?
@gwern@tenobrus Assuming they're currently going through a mess of hacker-adjacent intermediaries incentivized to mess things up worse than Anthropic can, data quality should be worth paying an Enterprise/Benchmark/Distillation-grade premium.
@gwern@tenobrus sure but every business has a range of fuckoff prices for unwanted customers and they *have* the CVP and similar pipelines in place
*shrug* being dramatic instead just seems unprofessional/unserious
@arkephi I think the Someone Should Do Something people (person, really) trying to turn it into a reputation risk for big labs kinda nuked it. We do need a softer, non-DSM term for harmless cranks getting LLM validation.
@LeroyLi311063 Nah.
Anthropic would love to, sure. But you think American feds tolerate weird cultists who can't shut up even when hundreds of millions of gov contracts depend on it?