https://x.com/justalexoki/status/1951410524996116681

That's all I got lol happy Friday

https://x.com/ycombinator/status/1951025508726239589

Actually here's some other dystopian shit. Fuck y combinator and fuck Asians and Indians

(http://www.autoadmit.com/thread.php?thread_id=5757240&forum_id=2:#49149741)

Language models sometimes go haywire and slip into weird and unsettling personas. Why? In a new paper, we find “persona vectors"—neural activity patterns controlling traits like evil, sycophancy, or hallucination.

https://x.com/AnthropicAI/status/1951317898313466361

Retroactively training artificial intelligence models backwards through time to be evil roko basilisks groyper

(http://www.autoadmit.com/thread.php?thread_id=5757240&forum_id=2:#49150080)

New paper: Since the rise of large language models, there's been a huge shift in academic writing.

In 2024, the word "delves" appeared 2,700% more than its historical average, by one account.

The analysis suggests that 13.5% of 2024 abstracts were processed with LLMs.

(http://www.autoadmit.com/thread.php?thread_id=5757240&forum_id=2:#49151190)

Machine learning pumo I need your hot take on this. What exactly is the mechanism causing this phenomenon? Is it just “evil” being less orthogonal than “good” relative to incorrect information and problem solving techniques as judged by the models existing weights? Is it incorrect facts and problem solving techniques generally accompanying surrounding “evil” inputs in the training data? What is causing this?

(http://www.autoadmit.com/thread.php?thread_id=5757240&forum_id=2:#49151205)