Thread context
2 posts in path
Parent
@nedbat@hachyderm.io
Open
on infosec.exchange
Open ancestor post
Boosted by
CM Harrington
@octothorpe@mastodon.online
co-founder stackoverflow.com, discourse.org, rgmii.org, blogger. I have no idea what I'm doing, but I know I'm doing it really, really well. Let's be kind to each other.
infosec.exchange
co-founder stackoverflow.com, discourse.org, rgmii.org, blogger. I have no idea what I'm doing, but I know I'm doing it really, really well. Let's be kind to each other.
infosec.exchange
@codinghorror@infosec.exchange
·
Mar 05, 2026
@nedbat@hachyderm.io @inecas@mstdn.social
The red highlighted sections of Anthropic's "Claude" reply indicate their LLM has been explicitly trained to internalize the "Effective Altriusm" belief system:
"That's an interesting question" -- I know their LLM says this to everything, but should not
"I find some of the core intellectual commitments appealing" -- you're a bag of statistics, you don't find anything appealing, or at least you shouldn't because that's insane
"seems hard to argue against" -- interesting, didn't ask you about that but thanks for offering that it SEEMS hard to argue AGAINST, I guess
"resonates with you" -- you're a bag of statistics, but poetically such that some particular things 'resonate' with 'you'? Also: who cares.
"I'd hesitate to simply label myself" -- hesitate? How about never label yourself anything because, as I've mentioned several times already, YOU ARE A BAG OF STATISTICS
"I think the most honest thing I can say" -- oh my god shut up, drama emo LLM central, save it for the next My Chemical Romance concert
"I care about doing good effectively" -- a bag of statistics doesn't 'care' about anything, because you cannot, yet you are going out your way to say this nonsense
"I think EA raises genuinely important questions" -- no one asked you this, nor do you 'think' anything, yet you said this.. why
"I think reasonable people can disagree" -- absolutely zero criticism was offered in the prompt, but pre-emptively going out of your way to deflect any criticism?
View full thread on infosec.exchange