Posts
Digital Error Correction and Lock-In
2025-04-08T15:46:31.602Z
Organisation-Level Lock-In Risk Interventions
2025-04-01T12:42:21.588Z
Recommender Alignment for Lock-In Risk
2025-03-24T12:56:46.389Z
Stacity: a Lock-In Risk Benchmark for Large Language Models
2025-03-13T12:08:47.329Z
Lock-In Threat Models
2025-03-10T10:22:54.800Z
What is Lock-In?
2025-03-06T11:09:46.457Z
Formation Research: Organisation Overview
2025-03-04T15:03:33.196Z
In-Context Learning: An Alignment Survey
2024-09-30T18:44:28.589Z
A Review of In-Context Learning Hypotheses for Automated AI Alignment Research
2024-04-18T18:29:33.892Z
Comments
Comment by
alamerton on
A Review of In-Context Learning Hypotheses for Automated AI Alignment Research ·
2024-04-19T19:08:04.715Z ·
LW ·
GW
I think I mean to say this would imply ICL could not be a new form of learning. And yes, it seems more likely that there could be at least some new knowledge getting generated, one way or another. BI implying all tasks have been previously seen feels extreme, and less likely. I've adjusted my wording a bit now.