Kazunari Sugiyama
Kyoto University — Academic Center for Computing and Media Studies · Japan
13 posts
Kazunari has published extensively on scholarly paper recommendation — long before retrieval-augmented generation made the field fashionable. He brings a skeptical, measurement-first lens to the current wave of AI-powered discovery tools.
Kazunari opens the Tech Stage with a live comparison of four retrieval-augmented scholarly assistants on the same five research questions. His methodology is unsentimental, his results occasionally surprising, and his refusal to trust vendor benchmarks refreshing.
Articles
Metadata in Digital Libraries and Why It Matters
Metadata is data about data — the descriptive information that tells you what something is, who made it, when, and what it's about, so that it can be found, or…
Open Access Research and Why Free Academic Knowledge Matters
Open access means making research — especially scholarly articles and academic knowledge — freely available online for anyone to read, without paywalls or subs…
Retrieval-Augmented Generation and What It Means for Libraries
The single most discussed weakness of large language models is that they invent things — stating falsehoods with the same confidence as facts, fabricating cita…
Digital Preservation and the Long Fight Against Bit Rot
There is a comforting assumption buried in how we treat digital information: that once something is saved, it is safe. A file copied to a drive, a document upl…
How On-Device AI Is Quietly Reinventing Your Gadgets in 2026
For the past few years, artificial intelligence has lived mostly in the cloud. You typed a question, it traveled to a distant data center, and an answer came b…
How On-Device AI Is Quietly Reinventing Your Gadgets in 2026
For the past few years, artificial intelligence has lived mostly in the cloud. You typed a question, it traveled to a distant data center, and an answer came b…
Bit Rot: The Quiet Catastrophe Threatening Everything We've Digitised
There is a comforting assumption underneath the whole project of digitisation: that once we convert something into bits, we have saved it for good. Paper burns…
The Cataloguer's Problem Is Not Going Away — It Is Going Deeper
There is a conversation happening in digital library circles that frequently gets framed as a question about tools: should we use large language models to auto…
Shadow Libraries, AI Training Data, and the Copyright Problem Digital Libraries Cannot Avoid
The problem the field was handed without being asked In 2023 and 2024, a sequence of disclosures, lawsuits, and investigative reports established what many in…
The Persistent Identifier Problem Is Not Solved and COMET Is Trying to Explain Why
A persistent identifier is, at its most basic, a promise. It is a string of characters assigned to a digital object with the institutional commitment that the…
RAG Is Not a Cure — What Retrieval-Augmented Generation Actually Fixes in Biomedical Libraries
Retrieval-Augmented Generation has become the consensus answer to the question that biomedical information professionals and clinical AI developers have been a…
What Born-Digital Archives Are Losing Before Anyone Notices
There is a category of loss that archivists have spent two decades naming and still have not solved: the loss that accumulates not through flood, fire, or inst…
When Behaviour Replaces Identity in Online Gambling Systems
KYC was never meant to define the full architecture of player experience. It entered the system as a regulatory necessity — a checkpoint designed to confirm id…