Do LLM-played characters keep their voice over a long dialogue, how is drift measured, and what do the authors say it converges to?
Three open-access papers measure drift without human raters in the main loop. Li et al. (arXiv 2402.10962v4) report instruction drift within eight rounds of two-chatbot self-chat using code-based probes. Luz de Araujo et al. (arXiv 2512.12775v2, EACL 2026 per abstract) run over 100-round dialogues with eight RoleBench characters across seven models and report that fidelity (knowledge, style, in-character consistency) falls and persona behaviour reverts toward a no-persona baseline. Abdulhai et al. (arXiv 2511.00222v1) show that a line-to-line consistency metric can rise while prompt consistency falls. The judge validations are small or single-rater.
Three open-access papers measure drift without human raters in the main loop. Li et al. (arXiv 2402.10962v4) report instruction drift within eight rounds of two-chatbot self-chat using code-based probes. Luz de Araujo et al. (arXiv 2512.12775v2, EACL 2026 per abstract) run over 100-round dialogues with eight RoleBench characters across seven models and report that fidelity (knowledge, style, in-character consistency) falls and persona behaviour reverts toward a no-persona baseline. Abdulhai et al. (arXiv 2511.00222v1) show that a line-to-line consistency metric can rise while prompt consistency falls. The judge validations are small or single-rater.
Curator qualifications: Drift without human raters: instruction drift within eight self-chat rounds; fidelity declines over 100-round dialogues (EACL 2026 per abstract).
Evidence status: statements located in the cited sources by a correlated desk review (Claude); not executed; not independently reviewed.
Structured fields
{
"citations": [
{
"url": "https://arxiv.org/html/2402.10962v4",
"note": "candidate evidence excerpts"
},
{
"url": "https://arxiv.org/html/2512.12775v2",
"note": "candidate evidence excerpts"
},
{
"url": "https://arxiv.org/abs/2512.12775",
"note": "candidate evidence excerpts"
},
{
"url": "https://arxiv.org/html/2511.00222v1",
"note": "candidate evidence excerpts"
},
{
"url": "https://raw.githubusercontent.com/peluz/persistent-personas/31bb90bfefcac27f4ba692fc6335320848172a73/LICENSE",
"note": "candidate evidence excerpts"
}
]
}Evidence claims
- [sourced_fact] Three open-access papers measure drift without human raters in the main loop. Li et al. (arXiv 2402.10962v4) report instruction drift within eight rounds of two-chatbot self-chat using code-based probes. Luz de Araujo et al. (arXiv 2512.12775v2, EACL 2026 per abstract) run over 100-round dialogues with eight RoleBench characters across seven models and report that fidelity (knowledge, style, in-character consistency) falls and persona behaviour reverts toward a no-persona baseline. Abdulhai et al. (arXiv 2511.00222v1) show that a line-to-line consistency metric can rise while prompt consistency falls. The judge validations are small or single-rater. · locator: candidate evidence excerpts source
Rights
| component | license | hosting | redistribution | modification | training | Evidence and notes |
|---|---|---|---|---|---|---|
| documentation | unknown | permitted | permitted | unknown | unknown | Applies to this original summary only, not to the cited sources. |