28 August 2026
From past chats
Keep an evidence ledger for claims about AI and work
Your earlier search for recent writing about AI and employees could become more useful as a small, durable evidence ledger rather than another pile of articles. Give each claim a row with the occupation or task studied, whether the result measures exposure, adoption, productivity, or actual job loss, the study population, and a review date. The important distinction is between a job containing automatable tasks and a worker actually being displaced. A good first experiment would be to score five memorable claims from recent newsletters against one primary study each; which claims survive when “exposure” and “employment effect” are kept separate?
- ILO: Generative AI and jobs — a 2025 update
- ILO: Generative AI at work — what it means for jobs
- ILO and World Bank: uneven global impact of generative AI on jobs
Give the RSS briefing a novelty test
The chat-history RSS briefing has accumulated good mechanical checks; the next useful measure is whether it earns attention. Make a frozen set of perhaps ten source snapshots and label each suggested topic as relevant, novel relative to the previous seven editions, and actionable. Run the current prompt and a deliberately simple baseline over the same set, then compare how many suggestions are both relevant and new. This borrows the shape of an information-retrieval test collection without needing a heavyweight evaluator, and it may reveal whether source selection or prose generation is the real bottleneck.
- NIST: How TREC builds reusable retrieval evaluations
- NIST: Overview of the TREC 2004 Novelty Track
- W3C Feed Validation Service
From the lmika.org front page
Separate today’s intentions from timed reminders
The daily handwritten list and Slack reminders are doing two different jobs, so it may help to make that boundary explicit. Keep the morning page as a disposable “today card” with one must-do item, two likely items, and a parking area for interruptions; use Slack only when the timing or message context matters. At shutdown, migrate rather than roll over: anything unfinished must be scheduled, returned to the longer-lived system, or deliberately dropped. After a week, ask whether the card reduced time spent choosing the next task without quietly becoming another backlog.
Listen to The Truman Show as two intertwined scores
Rather than simply putting the soundtrack back into rotation, try one focused pass that labels each cue as Burkhard Dallwitz, newly written Philip Glass, or an earlier Glass work repurposed for the film. Peter Weir’s notes describe the unusual conceit: some music belongs to the film, while some feels selected by Christof for the television show inside it. That makes a neat compositional study for your own game music—could one musical layer belong to the player’s world while another belongs to the system watching or directing it? Start by comparing “Truman Sleeps,” “It’s a Life,” and “Mishima / Opening,” then sketch a 30-second pair of cues with deliberately different authorship.