LESSWRONG
LW

3368
alamerton
49910
Message
Dialogue
Subscribe

Let your ideas die so you don't have to. Navigating parochialism. Fond of fallibilism.

I'm currently working as founder and research lead at Formation Research on technical interventions for lock-in risk, and part-time as a research assistant at King's College London on clinical machine learning benchmarking.

My website is here.

Sequences

Posts

Sorted by New

Wikitag Contributions

Comments

Sorted by
Newest
No wikitag contributions to display.
Lock-In
A Review of In-Context Learning Hypotheses for Automated AI Alignment Research
alamerton1y10

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.

Reply
1Digital Error Correction and Lock-In
6mo
0
5Organisation-Level Lock-In Risk Interventions
7mo
0
8Recommender Alignment for Lock-In Risk
7mo
0
4Stacity: a Lock-In Risk Benchmark for Large Language Models
7mo
0
5Lock-In Threat Models
7mo
0
5What is Lock-In?
7mo
0
5Formation Research: Organisation Overview
8mo
0
8In-Context Learning: An Alignment Survey
1y
0
25A Review of In-Context Learning Hypotheses for Automated AI Alignment Research
2y
4