corpus.blog
56,966 blogs · 4,417,066 posts Claim your blog
Record · corpus.blog/posts/01a0878e-c03c-708e-9d5a-cbbba37e3d8f

How human-like do safe AI motivations need to be?

Home - Joe Carlsmith · published 12 November 2025

A record is what was published and who pointed at it. The text of the post is not held here: read it at the source, or at its Wayback capture.

Read the post at joecarlsmith.com Wayback capture
3blogs citing
First cited
8 Dec 2025
Most recent
16 Mar 2026
Rank this month
not ranked
Links from this post
38
Words captured
16,612

Cited by

Blog
In the post
Date
zackmdavis.net
Terrified Comments on Corrigibility in Claude's Constitution Read at the source ↗
16 Mar 2026
forethought.org
Short Timelines Aren't Obviously Higher-Leverage Read at the source ↗
22 Jan 2026
gleech.org
AI in 2025: gestalt Read at the source ↗
8 Dec 2025

Links from this post

Link
Host
How do we solve the alignment problem? Read at the source ↗
joecarlsmith.com
https://www.amazon.com/Anyone-Builds-Everyone-Dies-Superhuman/dp/0316595640
amazon.com
Giving AIs safe motivations Read at the source ↗
joecarlsmith.com
Controlling the options AIs can pursue Read at the source ↗
joecarlsmith.com
AI for AI safety Read at the source ↗
joecarlsmith.com
Can we safely automate alignment research? Read at the source ↗
joecarlsmith.com
https://www.lesswrong.com/w/nearest-unblocked-strategy
lesswrong.com
Leaving Open Philanthropy, going to Anthropic Read at the source ↗
joecarlsmith.com
https://x.com/ESYudkowsky/status/1648518676084305926
x.com
https://en.wikipedia.org/wiki/Sydney_(Microsoft
en.wikipedia.org
https://www.anthropic.com/research/alignment-faking
anthropic.com
https://www.anthropic.com/research/agentic-misalignment
anthropic.com
https://arxiv.org/abs/2502.17424
arxiv.org
https://www.anthropic.com/research/persona-vectors
anthropic.com
https://palisaderesearch.org/blog/shutdown-resistance
palisaderesearch.org
Detecting and reducing scheming in AI models Read at the source ↗
openai.com
https://www.lesswrong.com/posts/qgvSMwRrdqoDMJJnD/towards-a-typology-of-strange-llm-chains-of-thought
lesswrong.com
https://www.arxiv.org/pdf/2509.15541
arxiv.org
https://www.antischeming.ai/cot-transcripts
antischeming.ai
https://www.antischeming.ai/cot-transcripts/figure-2-sandbag-model-graded-cot
antischeming.ai
https://www.lesswrong.com/posts/EaLCAZ4bgycY6hFC8/value-fragility-and-ai-takeover
lesswrong.com
An even deeper atheism Read at the source ↗
joecarlsmith.com
What is it to solve the alignment problem? Read at the source ↗
joecarlsmith.com
https://ifanyonebuildsit.com/11/shutdown-buttons-and-corrigibility
ifanyonebuildsit.com
https://www.lesswrong.com/posts/AFdvSBNgN2EkAsZZA/what-is-it-to-solve-the-alignment-problem-1
lesswrong.com
When should we worry about AI power-seeking? Read at the source ↗
joecarlsmith.com
https://arxiv.org/pdf/2311.08379
arxiv.org
https://www.lesswrong.com/posts/RQpNHSiWaXTvDxt6R/coherent-decisions-imply-consistent-utilities
lesswrong.com
Why should ethical anti-realists do ethics? Read at the source ↗
joecarlsmith.com
https://www.lesswrong.com/w/problem-of-fully-updated-deference
lesswrong.com
https://www.lesswrong.com/w/ontology-identification-problem
lesswrong.com
https://distill.pub/2017/feature-visualization/appendix/googlenet/4e.html
distill.pub
https://ifanyonebuildsit.com/4/if-current-ais-are-mostly-weird-in-extreme-cases-whats-the-problem
ifanyonebuildsit.com
Being nicer than Clippy Read at the source ↗
joecarlsmith.com
https://www.edge.org/conversation/the-myth-of-ai
edge.org
https://www.lesswrong.com/posts/2NncxDQ3KBDCxiJiP/cosmopolitan-values-don-t-come-free?commentId=sjutYfdA28BaovfPp
lesswrong.com
https://en.wikipedia.org/wiki/Coherent_extrapolated_volition
en.wikipedia.org
On the limits of idealized values Read at the source ↗
joecarlsmith.com