Experiment
Can AI write a modern jazz tune?
Claude and GPT coding agents each composed modern jazz tunes over two rounds. Listen, read the charts and vote before seeing the results.
Web apps, games, music tools and research side projects, mostly outside my work on The Unjournal. Many started as quick experiments built with AI coding agents. A few are now in regular use.
Tools for practising and studying jazz, plus the bands I play tuba and trumpet with.
App
Play and study 18 versions of the 12-bar blues in any key, from the basic three-chord form to Charlie Parker's 'Bird' changes.
Experiment
Claude and GPT coding agents each composed modern jazz tunes over two rounds. Listen, read the charts and vote before seeing the results.
App
Repertoire tracker, setlist builder, 12-key technique grid, practice-plan generator and metronome drills in one page.
Listings
Jazz concerts and jam sessions within about an hour of Monson or Framingham, Massachusetts.
Things for playing, borrowing and scheduling.
Gameno domain yet, plain HTTP
A multiplayer word game played in meaning-space. Add and subtract words to roll your ball toward a target word; word-vector similarity decides who is closest.
Search
Search about 2,000 tools, tech and gear you can borrow from public libraries and tool-lending libraries, starting in Massachusetts and spreading across the US.
Appprototype
Offer candidate times, share one link, watch responses come in and pick the best slot. Built for Unjournal team meetings.
App
A shared room where people say when they're free for a chat and how to reach them. No account needed; make a room for any group.
Short, sourced pitches for things I think someone should try. Feedback and collaborators welcome.
Proposalnew
Anonymous suggestion boxes backed by a reward pool, so organizations pay for the criticism that turns out to be useful.
Policy idea
A second look at my 2015 proposal for voluntary, income-verified discounts, with a research agenda for testing it.
Proposal
A testable design for training researchers around judgment and supervised practice, in a world where AI does more of the routine work.
Course
How language models work, for researchers who have to judge their output: tokens, measurement, evidence workflows, compute economics and safety.
Research tools and write-ups on effective giving, research evaluation and prioritization. Close to The Unjournal, but not part of its core tools.
Directory
Search organizations and products connected to substantial charitable giving. Every entry links to the original claim, its source and a confidence rating.
Paper and site
Co-authored study comparing one-shot LLM evaluations of social-science research with structured expert review.
Model
A sourced model and dashboard of how wealth from AI labs might, or might not, turn into philanthropic funding.
Interactive explainer
Simulations of preference learning, reward models, calibration and active learning, built while designing a research-prioritization pipeline.
Web book
The puzzle of why people don't give more effectively, what the evidence says, and what we still need to learn.
Web book
Field experiments and synthesis on whether telling donors about effectiveness changes what they give.
Reports
Public reports from a team running marketing and messaging trials with effective-giving organizations.
Open course notes and small methods tools, mostly from my university teaching.
Web book
Working notes with a micro, behavioral and experimental focus.
Web book
A practical guide to economics dissertations and research projects.
Calculator
Power analysis for comparing two independent groups with a Bayesian test.
The Unjournal's own tools live on their own page.
Tools hub
Evaluation forms, a claim highlighter, the research-prioritization dashboard, pivotal-questions workshops and models, and analysis tools for The Unjournal's evaluation work.