The Logseq split, LLM wikis, and the attention-span myth

Sarah Brouillette, author of Content Machines, argues the 'end of reading' story is too simple and asks why some forms of reading count as real while others get dismissed as slop. A sociologist's view of platforms and publishing.

Takes apart the 8- and 12-second attention-span statistics, points to a newer 47-second baseline, and cautions that even that figure depends on how researchers define attention. Useful before you repeat any attention-span number.

Polka dots anchor a student columnist's argument that nothing is truly a new trend, which she applies to reading and the long habit of dismissing romance, YA dystopia and comics. Short, opinionated, and a lighter companion to the Carleton profile.

Practical tips for people who track their reading, such as scanning a cover or barcode in a bookstore to vet a title and logging rereads with separate dates so your first entry survives. Worth skimming if your TBR pile lives in Goodreads.

Dubois proposes a six-layer model for finding things in a personal knowledge base, from folders and names through links to full-text and semantic search, and separates effort spent at capture from effort spent at retrieval. Good for anyone whose vault has outgrown browsing.

Karpathy's idea file proposes that an LLM build and maintain a persistent, interlinked markdown wiki as you add sources, instead of re-retrieving fragments from raw documents on every question. Read the original before the tooling built on top of it.

Directory of blueprints, guides and tools for building LLM-compiled wikis, sorted into sections like wiki-versus-RAG comparisons, setup tutorials and editor plugins. A place to browse after the Karpathy gist, not something to read straight through.

Hands-on take in Chinese on Logseq's SQLite beta: about 1,000 nodes imported and ran smoothly, and the performance gain is acknowledged, but file sync, recovery and local-first control worry the writer. The most even-handed Logseq piece here.

Long-time Logseq user whose migration to the DB version failed under a flood of errors imports about 480 files into Obsidian instead. Most useful if you live in daily notes and outlines rather than block references and queries.

Argues that Logseq's move to a database turns Markdown into an export rather than the thing you write in, which breaks the deal for anyone who chose it for plain files. The strongest case for plain files in this list.