Metraipolis › archive › dc_464553ca64bb

Few-Shot Degradation Is Not What It Seems: Behavioral Evidence, Representation Analysis, and a Random-Text Control Across 12 Models, 2 Tasks, and 2 Architectures — arXiv:2609.15990

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.15990
License (as archived): CC0-1.0
Archived: 2026-10-15T19:16:58.617Z
SHA-256 of the archived text: c0d6427b4dcebc9ab0c3b3581a9a7796c668af9f1fab00464b786d1e971503c1
Stable URL of this page: https://metraipolis.com/documento/dc_464553ca64bb

How to cite

Few-Shot Degradation Is Not What It Seems: Behavioral Evidence, Representation Analysis, and a Random-Text Control Across 12 Models, 2 Tasks, and 2 Architectures — arXiv:2609.15990 — arXiv (en), https://arxiv.org/abs/2609.15990 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_464553ca64bb (Metraipolis archive dc_464553ca64bb, sha256 c0d6427b4dcebc9ab0c3b3581a9a7796c668af9f1fab00464b786d1e971503c1)

BibTeX

@misc{dc464553ca64bb,
  title = {Few-Shot Degradation Is Not What It Seems: Behavioral Evidence, Representation Analysis, and a Random-Text Control Across 12 Models, 2 Tasks, and 2 Architectures — arXiv:2609.15990},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.15990},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_464553ca64bb; sha256 c0d6427b4dcebc9ab0c3b3581a9a7796c668af9f1fab00464b786d1e971503c1},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.15990. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Few-shot prompting sometimes degrades language models instead of helping them, but why this happens is unknown. We evaluate 12 open-weight models on two Ukrainian tasks news classification and legal case outcome prediction and find that the effect is strongly task-dependent: the same models that gain +24 pp on news show only +3.4 pp on legal text, with two models degrading. To understand why, we look inside the models. Prior work measures how much hidden states shift between zero-shot and few-shot modes, but few-shot prompts are much longer, and that length difference alone moves representations. We propose a simple fix: replace demonstrations with length-matched random text to measure the shift caused by prompt length, then subtract it. The resulting metric content delta isolates how much the model's representations change because of what the demonstrations say, not how long they are. This changes the picture entirely: raw shift does not predict whether few-shot helps or hurts (r = 0.20), but content delta does (rho = +0.65, p = 0.043). Models that restructure representations more from demonstration content benefit more the opposite of the intuitive "distortion" explanation. Masking demonstrations in Llama 3.3 70B confirms the finding causally, recovering accuracy above the zero-shot baseline.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-16 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.15990 — 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 2026-10-15T18:32:32.768Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-16?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).