2025-09-16
Tagged: cartesian tutor, llms
Progress update
Not much of note, website feature-wise. I finished my Stripe integration! (insert Scrooge duck diving into bathtub of gold meme.)
On the marketing front, I set up some semi-automated email scraping, where I looked into past winners of the USNCO, identified the schools they came from, figured out which chemistry teachers coached/advised them, and then looked up the teachers’ email addresses. The workflow was centralized around a single CSV file, to which I requested Claude do various lookups and to add in data. Unfortunately, Claude completely failed at reading some of the USNCO results PDFs, so I had to ask Gemini to transcribe them, and then dump the text into Claude to have it transfer it into the CSV. Claude also completely failed at web scraping emails, so I set up an MCP server wired up to Perplexity so that it would delegate its email scraping to Perplexity. So in the end, my system was a Frankenstein of Claude as Orchestrator, with Gemini as specialized image-transcriber and Perplexity as specialized web search agent. This system got me ~100 teacher emails, to which I will slowly trickle out some cold emails over the next week or two.
The email scraper I built might actually make for a decent single-purpose website that accepts CSV/excel/text files and for-loops over them to scrape emails for you. I’d guess it would take a week or two to set up properly, and would be an interesting exercise in configuring a production Claude orchestration agent. It would probably also make more money than the chemistry tutoring thing.
Taste
I wrote Taste to explore some of the swirling ideas in my head. Many ideas didn’t solidify in time for the Taste essay. These are my answer to the concluding question of that essay.
- When is a taste-based product or service profitable? Is profit necessarily in opposition to taste? What systems of governance align these two incentives?
- What is the nature of the interaction between community, channels, and platforms? Why have Bluesky, Lobste.rs, Nebula not really take off? Why is Substack seemingly succeeding (and how did Medium screw it up so badly)? Is Youtube’s advantage in its creators, rather than its algorithm?
- Are there algorithmic solutions (a la Pagerank) that assign “taste” scores to users as well as content? Can these algorithmic solutions bootstrap good taste measurement abilities, or do low-taste equilibria dominate? Perhaps early Google was successful precisely because PageRank distilled the good taste of a very small, elite group of folks that populated the early Internet, but everything’s been downhill since then, given the parasitic pressure of SEO optimizers and eternal September of tasteless new people on the internet.
- What is the process by which low-taste communities slowly bootstrap themselves into a high-taste community? How does the pace of change, and parasitic load of grifters affect how and whether this transition ever happens?