Newest papers

What Language Adds, and at What Cost: Four Controlled Probes of Pathology Vision-Language Models

Wenhao Zhang, Zhongliang Zhou, Yi Ping, Zirui Wang, Sheng Li · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26805 · cited by 0 works (OpenAlex) · open-access PDF

Indonesian

Seberapa banyak bahasa menambahkan di luar tulang belakang penglihatan tidak jelas, karena hampir setiap sistem dievaluasi oleh para pengembang pada set pengujian pilihan mereka sendiri. Kami menilai bukti yang dipublikasikan oleh bagaimana itu diproduksi dan menguji empat klaim dengan prob yang dikendalikan yang menggunakan kembali checkpoints yang dilepaskan dan data publik, bervariasi satu input pada satu waktu. Teks laporan menambahkan sinyal prognostik: menambahkan embedding beku dari laporan yang diedit tahap ke embedding slide meningkatkan indeks C dikumpulkan dengan 0,019 untuk interval progresi bebas dan dengan 0,016 untuk keseluruhan kelangsungan hidup di 1{33}6 pasien TCGA, dan mengembalikan pernyataan tahap menambahkan. Prompts menambahkan tidak ada akurasi yang dapat diukur pada kesimpulan: dengan menggabungkan fitur dan modwers fixed, klasifikasi nol-shot secara statistik

Machine translation (Meta M2M100); the English original follows.

Abstract

Vision--language models (VLMs) promise to make computational pathology queryable and able to draft reports, but how much language adds beyond the vision backbone is unclear, because almost every system is evaluated by its developers on test sets of their own choosing. We grade published evidence by how it was produced and test four claims with controlled probes that reuse released checkpoints and public data, varying one input at a time. Report text adds prognostic signal: adding a frozen embedding of the stage-redacted report to a slide embedding raises the pooled C-index by 0.019 for progression-free interval and by 0.016 for overall survival across 1{,}633 TCGA patients, and restoring the stage statement adds nothing. Prompts add no measurable accuracy at inference: with features and pooling fixed, zero-shot classification is statistically indistinguishable from a supervised linear probe for lung and renal cancer subtyping. Slide-level question answering is barely grounded: SlideChat answers 74.82\% of SlideBench-VQA(TCGA) items correctly with the slide and 74.41\% without it, whereas its language backbone alone reaches 43.52\%. Paraphrasing leaves accuracy unchanged but flips 8--23\% of answers. Self-reported gains from agentic navigation and few-shot language priors cannot yet be attributed to language. Language thus adds real but narrow value, and we propose minimum controls that would expose these failure modes in future evaluations.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Pooling, Computer science, Natural language processing, Reuse, Embedding, Artificial intelligence

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange