Metraipolis › archive › dc_f421bbcc96c5

Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs — arXiv:2609.29845

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.29845
License (as archived)
CC0-1.0
Archived
2026-10-08T04:27:48.754Z
SHA-256 of the archived text
e0de2e2178d1fb39e85de2cce5472da4aca309e6963b2f2c75ecfe8a6e338d11
Stable URL of this page
https://metraipolis.com/documento/dc_f421bbcc96c5

How to cite

Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs — arXiv:2609.29845 — arXiv (en), https://arxiv.org/abs/2609.29845 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f421bbcc96c5 (Metraipolis archive dc_f421bbcc96c5, sha256 e0de2e2178d1fb39e85de2cce5472da4aca309e6963b2f2c75ecfe8a6e338d11)
BibTeX
@misc{dcf421bbcc96c5,
  title = {Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs — arXiv:2609.29845},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.29845},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f421bbcc96c5; sha256 e0de2e2178d1fb39e85de2cce5472da4aca309e6963b2f2c75ecfe8a6e338d11},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.29845. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

While Large Language Models (LLMs) rely on highly non-linear components, in this work we demonstrate that they exhibit fundamental linearity: when inputs from distinct text streams are linearly combined, the model outputs a superposition of the individual next-token distributions. We term this the \textit{Superposition Linearity Hypothesis}. We provide evidence that superposition is an intrinsic property of the Transformer architecture rather than an emergent consequence of training; in fact, we observe that it tends to diminish as pretraining progresses. However, we demonstrate that linearity can be substantially restored through lightweight fine-tuning, significantly reducing the divergence between the predicted next-token distribution and the average of the individual next-token distributions. Finally, we introduce a guided decoding procedure that disentangles superposed outputs, enabling the simultaneous generation of two coherent continuations from a single forward pass.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-25 in cs.AI).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.29845 — 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 2026-10-08T03:50:58.170Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-25?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).