It's a symbiotic relationship: AI doomers need AI hype in order to be taken seriously and be part of the discourse, and AI hypers need AI doom narratives for AI to look momentous and world-altering.
SAFETY
-
AI Mental Health Systems: Data Privacy and APA Compliance Measures
By
–
How are they protecting prisoner’s data? And what measures are the psychiatrist(s) taking to ensure they’re following mental health guidelines from APA or other trusted orgs? And thanks.
-
Universal Adversarial Attacks on Aligned Language Models Study
By
–
Universal and Transferable Adversarial Attacks on Aligned Language Models by @andyzou_jiaming et al. is wild. “Adversarial prompts generated by our approach are quite transferable, including to black-box, publicly released LLMs.” https://
arxiv.org/abs/2307.15043 https://
github.com/llm-attacks/ll
m-attacks
… -
Favorite Discussion on LLM Jailbreaks and Adversarial Attacks
By
–
My favorite part is the discussion about jailbreaks and adversarial attacks on LLMs near the end!
-
Should AI Weight Known False Evidence in Reasoning?
By
–
Do you give 10% weight to a proof that you know is wrong?
-
Collective Defense Against Existential AI Risks
By
–
The answer is no, with any budget.
And even if it were possible for someone to do this, it would also be possible for a bunch of people to design a system to take down the world ender. -
Flawed Logic in Scientific Opinion Formation and Consensus
By
–
I'd like to respectfully point out that the logic in this argument is based on a flawed model for how scientists think. Scientists don't just take a weighted average of others' opinions to form their own. A good scientist takes as input lots of data, including others' opinions,
-
Reinforcement Learning Applied to Encrypted Data Security
By
–
Reinforcement Learning on Encrypted Data Jesu et al.: https://
arxiv.org/abs/2109.08236 #ReinforcementLearning #CyberSecurity #QLearning -
P(doom) Metrics Meaningless: Always Trending Toward 100 Percent
By
–
The percentage given for P(doom) is essentially meaningless – it's always heading towards 100%, much like a loading bar.