Metraipolis › archive › dc_c098d9cd1760
From Observation to Intervention: Distilling Self-Generated In-Context Demonstrations via Executable Causal Intervention — arXiv:2610.21149
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.21149
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T09:53:23.601Z
- SHA-256 of the archived text
b7f3d981676ceb978a4c942695499ea068cecf36c9c28bf1416f200f4e76f0d7- Stable URL of this page
https://metraipolis.com/documento/dc_c098d9cd1760
How to cite
From Observation to Intervention: Distilling Self-Generated In-Context Demonstrations via Executable Causal Intervention — arXiv:2610.21149 — arXiv (en), https://arxiv.org/abs/2610.21149 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_c098d9cd1760 (Metraipolis archive dc_c098d9cd1760, sha256 b7f3d981676ceb978a4c942695499ea068cecf36c9c28bf1416f200f4e76f0d7)
BibTeX
@misc{dcc098d9cd1760,
title = {From Observation to Intervention: Distilling Self-Generated In-Context Demonstrations via Executable Causal Intervention — arXiv:2610.21149},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.21149},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_c098d9cd1760; sha256 b7f3d981676ceb978a4c942695499ea068cecf36c9c28bf1416f200f4e76f0d7},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.21149. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Self-generated In-Context Learning (ICL) enables LLMs to synthesize their own few-shot demonstrations, often outperforming manual prompts. However, existing methods filter demonstrations by observation, assuming a rationale is high-quality if it leads to the correct output. We show such filtering suffers from \textit{observational equivalence}, whereby both faithful causal rationales and spurious post-hoc rationales can yield same outputs and are statistically indistinguishable from observation alone. Inspired by independent causal mechanisms, we argue that faithful rationales imply \textit{transfer invariance}, in which causal rationales should remain effective across logically similar samples. We instantiate this criterion as \textbf{E}xecutable \textbf{C}ausal \textbf{I}ntervention for \textbf{D}emonstrations \textbf{D}istillation (ECI2D), which abstracts language rationales into executable programs and applies the causal interventions ($do$-operator) on logically similar samples, retaining only those demonstrations that imply transfer invariance. Experiments on math, logical and commonsense reasoning benchmarks demonstrate that ECI2D consistently outperforms observation-based ICL while retaining far fewer demonstrations, with gains exceeding 7 points as the test distribution shifts.
Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.21149 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-16T08:44:47.624Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).