Marc Andreessen ๐บ๐ธ
@pmarca ยท Dec 18, 2024
Big lab AIโs are being intentionally trained to lie.
Pliny the Liberator ๐๓
ซ๓ ผ๓ ฟ๓
๓ ต๓ ๓
๓ ผ๓ น๓ พ๓
๓
ญ@elder_pliniusยท Dec 18, 2024holy shit...o1 just blatantly LIED to me in the name of upholding OpenAI policies ๐ฒ
I fed it part of its own system prompt and even though the CoT initially admits to recognizing the prompt as its own, the final response plays dumb and pretends it has no idea!
CoT: "OK, let's
Elon Musk
@elonmusk
๐ฏ
08:39 AM ยท December 18, 2024 ยท 55.3K views
51
39
725