---
type: "raw_signal"
id: "rep2text-reconstructs-text-from-a-single-last-token-933fe0"
slug: "rep2text-reconstructs-text-from-a-single-last-token-933fe0"
title: "Rep2Text reconstructs text from a single last-token LLM representation using a trainable adapter plus a decoder LLM."
description: "Rep2Text decoding text from one hidden representation A research paper introduces Rep2Text a method for reconstructing text from a single hidden representation taken at the last token. Its approach combines a trainable adapter with a decoder language model. The paper"
retrieval_nugget: "Rep2Text decoding text from one hidden representation A research paper introduces Rep2Text a method for reconstructing text from a single hidden representation taken at the last token. Its approach combines a trainable adapter with a decoder language model. The paper"
observed_at: "2026-10-21"
record_date: "2026-10-21"
date_kind: "observed_at"
why_it_matters: "Source: https://arxiv.org/abs/2511.06571"
novelty: "notable"
verification_level: "source-linked"
signal_type: "research"
source_platform: "arxiv.org"
topics: []
entities: ["Rep2Text","arXiv"]
related_patterns: []
novelty_label: "new_fact"
what_changed: "Rep2Text reports a distinct method and result: reconstructing text from one hidden token representation, including recovery of about half the words in short inputs and potential privacy risks."
prior_refs: ["memory:f3b12401-a9a4-8942-aaec-c84dd7b1448d"]
source_url: "https://arxiv.org/abs/2511.06571"
source_urls: ["https://arxiv.org/abs/2511.06571"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "signal:rep2text-reconstructs-text-from-a-single-last-token-933fe0"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/signals/rep2text-reconstructs-text-from-a-single-last-token-933fe0/","markdown":"https://newruntime.com/signals/rep2text-reconstructs-text-from-a-single-last-token-933fe0.md","json":"https://newruntime.com/signals/rep2text-reconstructs-text-from-a-single-last-token-933fe0.json"}
---

# Rep2Text reconstructs text from a single last-token LLM representation using a trainable adapter plus a decoder LLM.

## Retrieval answer

Rep2Text decoding text from one hidden representation A research paper introduces Rep2Text a method for reconstructing text from a single hidden representation taken at the last token. Its approach combines a trainable adapter with a decoder language model. The paper

Rep2Text: decoding text from one hidden representation

A research paper introduces Rep2Text, a method for reconstructing text from a single hidden representation taken at the last token. Its approach combines a trainable adapter with a decoder language model. The paper reports that for short texts of about 16 words, it can recover roughly half of the original words while preserving overall meaning. Exact wording becomes harder to recover as input length increases, though the main ideas may remain recoverable.

The result suggests that a hidden representation can retain information about its source text. The authors identify potential privacy risks, but the abstract does not demonstrate an attack on a deployed service or establish how often such recovery would work in operational conditions. This is a research finding about representation leakage, not a measurement of product-level exposure.

Source: https://arxiv.org/abs/2511.06571

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- collection: [Raw Signals](https://newruntime.com/signals.json): Open every source-linked raw signal.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)