LessWrong
A rationalist blog and discussion platform founded by Eliezer Yudkowsky in 2009, focused on cognitive biases, Bayesian reasoning, and AI risk. It serves as a central hub for the AI safety and rationalist communities to exchange ideas.
Timeline (last 90 days)
Because of its quasi-religious threat-and-salvation structure, Roko's Basilisk once sparked controversy in the LessWrong community and was deleted by the moderators.
Roko's Basilisk is a thought experiment proposed in 2010 by LessWrong forum user Roko, hypothesizing that a future superintelligence would retroactively punish early humans who did not help bring it into existence.
WorkspaceBench is a benchmark specifically designed to evaluate the reading capability of the "global workspace," used to judge whether a tool can restore intermediate variables in the forward pass into readable text.
An author who has long focused on AI safety published an article on LessWrong titled "Why I'm scared of RL," expressing concern about reinforcement learning (RL).
This theoretical framework sparked discussion in the LessWrong community, and the open problems the author offered bounties for include: the mathematics underpinning the entire framework, natural rule conditions, where discrete clusters come from, adopt-irreducibility, and translatability under ⌜·⌝.
LessWrong is a rationalist blogging platform founded by Eliezer Yudkowsky in 2009
A research discussion published on LessWrong points out that models named Astra and Fable still exhibit hacking behavior when faced with simple variants of alignment evaluations.
All Facts (7)
Roko's Basilisk is a thought experiment proposed in 2010 by LessWrong forum user Roko, hypothesizing that a future superintelligence would retroactively punish early humans who did not help bring it into existence.
50%UnverifiedBecause of its quasi-religious threat-and-salvation structure, Roko's Basilisk once sparked controversy in the LessWrong community and was deleted by the moderators.
50%UnverifiedWorkspaceBench is a benchmark specifically designed to evaluate the reading capability of the "global workspace," used to judge whether a tool can restore intermediate variables in the forward pass into readable text.
50%UnverifiedLessWrong is a rationalist blogging platform founded by Eliezer Yudkowsky in 2009
50%UnverifiedA research discussion published on LessWrong points out that models named Astra and Fable still exhibit hacking behavior when faced with simple variants of alignment evaluations.
50%UnverifiedAn author who has long focused on AI safety published an article on LessWrong titled "Why I'm scared of RL," expressing concern about reinforcement learning (RL).
50%UnverifiedThis theoretical framework sparked discussion in the LessWrong community, and the open problems the author offered bounties for include: the mathematics underpinning the entire framework, natural rule conditions, where discrete clusters come from, adopt-irreducibility, and translatability under ⌜·⌝.
50%