I’m not arguing against those, at this point. But, what do you say to someone who has cancer or schizophrenia (etc.) and will die in a few months, if no new good cure is found? Last hope is fast new discoveries here. Why would you delay that, for arguments based on lies to
SAFETY
-
Utilitarian AI Ethics: Addressing Historical Harms and Future Risks
By
–
Your notes justly point out the amazing things done by the first, without grappling with the major problems of the 2nd, or with the historical horrors of unfettered utilitarianism in general.
-
Artificial General Intelligence Is Nigh: Rejoice, Be Very Afraid
By
–
h/t @RnA_EA
, https://
ea.rna.nl/2023/11/26/art
ificial-general-intelligence-is-nigh-rejoice-be-very-afraid/
… -
LLM Hallucinations Versus Fallible Memory: Key Distinction
By
–
Fallible memory ≠ hallucination. When an LLM says Elon Musk died in a car accident, it is generating something that never happened & failing to validate against immense evidence that he is still alive. Totally different phenomenon from you forgetting where you left your keys.
-
Buster Framework Combats Misinformation with Accurate Data Retrieval
By
–
This tech innovation uses Buster, an open-source framework, to embed and retrieve data, ensuring accurate, current responses, and combating misinformation.
-
AGI Ruin: A Comprehensive List of Lethalities and Counterarguments
By
–
It's not great, but it sure beats the detail levels of any counterclaims: https://
lesswrong.com/posts/uMQ3cqWD
PHhjtiesc/agi-ruin-a-list-of-lethalities
… -
Heat Transfer Limitations in Advanced Computational Systems
By
–
Terrible heat transfer. Dumping all your entropy into Earth's oceans as a negentropy sink doesn't last but it gets you a valuable early burst of computation before you start disassembling planets to enclose the Sun in solar panels, which Earth also doesn't survive.
-
Safety Plan Criticism: Promise Not Policy
By
–
That’s not a safety plan, nor a policy plan. It’s a promissory note. https://
x.com/ylecun/status/
/ylecun/status/1728841630709645379
… -
Instrumental Objectives and Low-Level Guardrails in AI Design
By
–
Right, the old "instrumental objective" story.
You can have low-level guardrails against bad effects of instrumental sub-goals.
The question here is not "can you come up with a way that this could go wrong?", but rather "is there a way to do it right?" It's like turbojet design.