Missed Connections and Markov Chains (nry.me)

A nostalgia-driven developer resurrected a very old-school generative approach — a Markov chain — by feeding it years of Craigslist “missed connections” scraped weekly via a cron job and stored in SQLite. The chain generates short, often quirky posts that the author automatically publishes to Bluesky and Nostr, and because the source data is brief and abundant, the simple probabilistic model produces a surprisingly high hit rate for believable, human-sounding snippets (examples include brief, oddly specific lines about weight, clothing, and casual encounters). The project underscores two points relevant to the AI/ML community: first, simpler models can still be creative and useful when matched to the right domain and corpus; second, there’s cultural value in the unpredictability and imperfections that modern, heavily curated LLM outputs tend to smooth away through attention architectures and RLHF. Technically, it’s a lightweight pipeline — periodic scraping → SQLite corpus → Markov-chain generator → social posting — that highlights how minimal infrastructure plus abundant short-form text can yield compelling generative behavior. The experiment is a reminder that not all progress must aim for polished perfection: sometimes the rough edges are where surprise and delight live.