Braille literacy among blind school-age children fell from roughly 50 percent in 1960 to under 10 percent today, and the decline is a content problem, not a technology problem: producing one illustrated braille page still requires specialized software and trained labor, so almost nothing in the existing catalog is personalized to an individual child. We present Tact, a system that converts a spoken or typed story idea into a printable braille page with a matching raised tactile illustration, computed inside a web browser with no account and no mandatory cost, and able to run with no server at all. This paper documents the full engineering history behind that system, rather than a narrow slice of it. It covers the target population and the ethical framing behind an explicit sighted-operator model, the hardware rationale for desktop fused deposition modeling over purpose-built embossing, and the physical braille geometry and its printer-specific calibration. It documents the architecture that lets a language model run on the reader’s own device, the later addition of a faster hosted path in front of it after an independent reimplementation exposed the cost of a 1.8 gigabyte first-use download, and the fallback chain that keeps the system working when that path is absent, the investigation that replaced a general-purpose braille translation library with a small deterministic implementation after it failed in production, and the evolution of the page layout from an inherited portrait specification to a verified square design, and its later revision from the field feedback of a teacher of blind readers. It further documents the construction of a 93-shape hand-drawn tactile illustration library derived from a frequency study of fairy-tale subjects, and the synthesized sound design used to make a voice-first interface legible without a screen. We report the verification performed to date, the ethical commitments the project has made, and the work required before the system can be considered ready for real blind and low-vision readers.
Retrieval-augmented generation (RAG) grounds a language model in retrieved documents, which reduces hallucination but creates a new attack surface: if retrieved text is tampered with, the model may repeat the falsehood. We study how much a small quantized model, Llama 3.1 8B, degrades when a fraction of its retrieved context is poisoned. Three corruption strategies are tested, entity swap, number swap, and negation, each applied to zero, one, two, or three of the three retrieved passages, over a factorial sweep of 588 runs on a fact-checking task built from FEVER. Accuracy falls from 77.9% on clean context to 43.5% when all three passages are corrupted. Entity swap flips the largest share of answers that were correct on clean context. Number-based corruption stays flat while poisoned passages are a minority and jumps once they form a majority, a pattern we re-check with query-level bootstrap intervals. The model rarely invents new falsehoods; its dominant reaction is to abstain, and a lexical overlap proxy of unsupported generation falls under attack rather than rising. The study is a small-scale measurement with coarse automated labels; we treat the strategy contrasts as suggestive until decoding is controlled and stronger adjudication is in place.
Braille literacy among blind school-age children has fallen sharply, in part because producing illustrated braille pages still requires specialized software and trained labor. We present Tact, a browser-based pipeline that converts a spoken or typed story idea into printable braille with a matching raised tactile illustration, without an account or mandatory cost and with an offline-capable path. The paper documents the engineering history of the system: its sighted-operator ethical model; hardware rationale for consumer fused-deposition modeling; physical braille geometry and printer calibration; local, hosted, and fallback language-model paths; a deterministic Grade 1 braille translator; verified page layout and pagination; a 93-shape hand-drawn tactile illustration library; and synthesized sound design for a voice-first interface. We report engineering verification, ethical commitments, limitations, and the work required before the system is ready for real blind and low-vision readers.
Iliano Fasolino· Zenodo (CERN European Organi...· 0 citations
Retrieval-augmented generation (RAG) grounds a language model in retrieved documents, which reduces hallucination but creates a new attack surface: if retrieved text is tampered with, the model may repeat the falsehood. We study how much a small quantized model, Llama 3.1 8B, degrades when a fraction of its retrieved context is poisoned. Three corruption strategies are tested, entity swap, number swap, and negation, each applied to zero, one, two, or three of the three retrieved passages, over a factorial sweep of 588 runs on a fact-checking task built from FEVER. Accuracy falls from 77.9% on clean context to 43.5% when all three passages are corrupted. Entity swap flips the largest share of answers that were correct on clean context. Number-based corruption stays flat while poisoned passages are a minority and jumps once they form a majority, a pattern we re-check with query-level bootstrap intervals. The model rarely invents new falsehoods; its dominant reaction is to abstain, and a lexical overlap proxy of unsupported generation falls under attack rather than rising. These results quantify a practical weakness of RAG and identify abstention, not fabrication, as the main behaviour to plan for. The study is a small-scale measurement with coarse automated labels; we treat the strategy contrasts as suggestive until decoding is controlled and stronger adjudication is in place.
Iliano Fasolino· Zenodo (CERN European Organi...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
