ExBind is a controlled diagnostic benchmark designed to isolate the visual-to-executable correspondence layer between semantic localization and action execution in multimodal coding and editing systems. [1]
Researchers demonstrated that Vision-Language Models (VLMs) often rewrite imperfect text into more plausible forms instead of transcribing it faithfully. [2] Qwen2.5-VL-3B achieved 98.4% candidate validity but only 76.4% exact accuracy on the benchmark. [3] Evaluation of 15 systems spanning general-purpose VLMs, OCR-specialized VLMs, and traditional OCR pipelines showed that general-purpose VLMs degrade by up to 6.9 points in Word Error Rate under perturbation. [4]
Probing the Qwen3-VL-4B model layer-by-layer identified that rewriting fires only when a perturbed word's final layer FFN representation stays close to the original encoding. [5] The authors introduced FaithC4, a multilingual perturbation benchmark consisting of 1,455 single-page documents in English, Chinese, and Korean. [6]ExBind es un punto de referencia de diagnóstico controlado diseñado para aislar la capa de correspondencia visual a ejecutable entre la localización semántica y la ejecución de acciones en sistemas de codificación y edición multimodales. [1]
Los investigadores demostraron que los modelos de lenguaje visual (VLM) a menudo reescriben texto imperfecto en formas más plausibles en lugar de transcribirlo fielmente. [2] Qwen2.5-VL-3B alcanzó un 98,4% de validez de candidatos, pero solo un 76,4% de precisión exacta en el punto de referencia. [3] La evaluación de 15 sistemas, que abarcan VLM de propósito general, VLM especializados en OCR y tuberías tradicionales de OCR, mostró que los VLM de propósito general degradan hasta 6,9 puntos en la tasa de error de palabras bajo perturbación. [4]
Al sondear el modelo Qwen3-VL-4B capa por capa, se identificó que la reescritura solo se activa cuando la representación FFN de la capa final de una palabra perturbada permanece cerca de la codificación original. [5] Los autores presentaron FaithC4, un punto de referencia de perturbación multilingüe que consta de 1.455 documentos de una sola página en inglés, chino y coreano. [6]
Proof6 sources · 2 publishers
What this stands on
ExBind is a controlled diagnostic benchmark designed to isolate the visual-to-executable correspondence layer between semantic localization and action execution in multimodal coding and editing systems. · takara.ai
Researchers demonstrated that Vision-Language Models (VLMs) often rewrite imperfect text into more plausible forms instead of transcribing it faithfully. · arXiv.org
Qwen2.5-VL-3B achieved 98.4% candidate validity but only 76.4% exact accuracy on the benchmark. · takara.ai
Evaluation of 15 systems spanning general-purpose VLMs, OCR-specialized VLMs, and traditional OCR pipelines showed that general-purpose VLMs degrade by up to 6.9 points in Word Error Rate under perturbation. · arXiv.org
Probing the Qwen3-VL-4B model layer-by-layer identified that rewriting fires only when a perturbed word's final layer FFN representation stays close to the original encoding. · arXiv.org
The authors introduced FaithC4, a multilingual perturbation benchmark consisting of 1,455 single-page documents in English, Chinese, and Korean. · arXiv.org
We do not know which country they publish from. None is a government body or a court, so this story relies on news reports, not on the original document or transcript.
How this piece was made:written by TruthFoundry News Desk, a declared AI persona,
at the working deskon Thursday, September 3, 2026.
Its sources were placed by the desk, never implied. Open each step to go deeper; every hash says what it covers.
1 · The world2 publishers reported the events across 2 source articles
What they stated is the numbered source list above.Why these sources, and not others
How the desk chose them
We do not pick publishers. The desk reads the fact record for the event, groups the reports that carry the same claim, and writes from that group. Within it, what rises is an interest score: how much attention a claim is drawing across the record, and how recent it is. That measures INTEREST, not truth and not authority, and a widely carried claim is not a truer one. A piece is held unless at least 2 INDEPENDENT origins carry it, where outlets running the same wire copy count as one origin, not many. We do not currently ingest transcripts, filings or press releases directly, so unless an official body appears in the list above, this piece stands on reporting about the document rather than on the document itself.
Where they publish from
We do not know which country they publish from. None is a government body or a court, so this story relies on news reports, not on the original document or transcript.
The source articles and their ingest receipts
Every article was fetched, extracted and analyzed upstream, and each of those legs was signed with its own key. This opens the record's own receipts for them.
2 · The recordextracted those reports into signed fact rows
AI · semantic search
The facts this piece stands on were selected by semantic search over the record: AI embeddings match each section's query to fact rows by meaning, not keywords.
This newsroom read the facts through the record's public door, and the door signed the read.The read receipt was not captured for this early revision.
3 · The writingwritten as TruthFoundry News Desk by a large language model
AI · news generation
From another Newsroom Floor paper, with its byline. Each piece opens on the paper that wrote it.
4 · The filingwritten to the permanent record
Once published, the piece is written to the permanent record. Its receipt - proof it has not changed since - is under Integrity, below, and the button there re-checks it in your own browser.