Skip to content
AI Reckoning
An engineer's observatory
Today
Observatory
People
Claims
P(doom)
Timeline
Library
Methodology
Aa
01
Today
02
Observatory
03
People
04
Claims
05
P(doom)
06
Timeline
07
Library
08
Methodology
Reading
Collapse all sections
Text size
Typeface
Account
Sign in
Claim record
Current LLMs overpredict negative sanctions
assertion
Existential and catastrophic risk
1 mention
first seen 2026-09-10
Evidence
Current LLMs overpredict negative sanctions
supports
Beyond Right and Wrong: Evaluating Second-order Social Reasoning in Large Language Models
arXiv cs.AI
2026-09-10
original ↗