This describes using Meta’s Muse not as a personal assistant but as a brute-force web-scraping workhorse to find and vet concert videos for a side project called fullsets.fm, which aggregates professionally recorded performances and extracts tracklists. The scraper pipeline uses Luna and Gemini 3.8 (via Gemini Studio) to cheaply assess audio/video quality and content, but discovery across many small YouTube channels and Reddit required a different tool. Muse can be instructed to scan thousands of videos or iterate artist-by-artist, running many parallel browser processes on a managed VM with Chrome to hunt down rare uploads and produce candidate lists; it also helps with artist matching, reviewing Luna/Gemini outputs, and final accept/reject decisions.
Practical findings: a month of Muse cost $80 and the system consumed billions of tokens per week, often seeming not to count usage; Muse Spark 1.3 is mediocre at general writing and coding but handles this large-scale, Gemini-Flash-style workload well, effectively offering near-unlimited tokens and cloud CPU/bandwidth. Downsides include job freezes, clunky multi-job UI, occasional frustrating model communication, and an obvious abuse vector that has already led to blocks (Amazon), raising concerns about future site defenses. A referral code (69WSDQ) was offered for signup bonuses.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.