Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Lucassen, James, Henry, Mark, Wright, Philippa, Yeung, Owen
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2408.15116
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910579370754048
author Lucassen, James
Henry, Mark
Wright, Philippa
Yeung, Owen
author_facet Lucassen, James
Henry, Mark
Wright, Philippa
Yeung, Owen
contents Many theoretical obstacles to AI alignment are consequences of reflective stability - the problem of designing alignment mechanisms that the AI would not disable if given the option. However, problems stemming from reflective stability are not obviously present in current LLMs, leading to disagreement over whether they will need to be solved to enable safe delegation of cognitive labor. In this paper, we propose Counterfactual Priority Change (CPC) destabilization as a mechanism by which reflective stability problems may arise in future LLMs. We describe two risk factors for CPC-destabilization: 1) CPC-based stepping back and 2) preference instability. We develop preliminary evaluations for each of these risk factors, and apply them to frontier LLMs. Our findings indicate that in current LLMs, increased scale and capability are associated with increases in both CPC-based stepping back and preference instability, suggesting that CPC-destabilization may cause reflective stability problems in future LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2408_15116
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Evaluating Stability of Unreflective Alignment
Lucassen, James
Henry, Mark
Wright, Philippa
Yeung, Owen
Artificial Intelligence
Many theoretical obstacles to AI alignment are consequences of reflective stability - the problem of designing alignment mechanisms that the AI would not disable if given the option. However, problems stemming from reflective stability are not obviously present in current LLMs, leading to disagreement over whether they will need to be solved to enable safe delegation of cognitive labor. In this paper, we propose Counterfactual Priority Change (CPC) destabilization as a mechanism by which reflective stability problems may arise in future LLMs. We describe two risk factors for CPC-destabilization: 1) CPC-based stepping back and 2) preference instability. We develop preliminary evaluations for each of these risk factors, and apply them to frontier LLMs. Our findings indicate that in current LLMs, increased scale and capability are associated with increases in both CPC-based stepping back and preference instability, suggesting that CPC-destabilization may cause reflective stability problems in future LLMs.
title Evaluating Stability of Unreflective Alignment
topic Artificial Intelligence
url https://arxiv.org/abs/2408.15116