Mike Noe

I added prompt caching to my Anthropic Batch API workflow. The hit rate was 0%.

Each model has a minimum cacheable token count — 4,096 for Haiku 4.5. If your cache_control block is below that, the API silently ignores it. Successful response, zero cache reads, no warning.

My IAB taxonomy prompt was 1,064 tokens. Well under the threshold.

Full write-up:

https://mikenoe.com/posts/prompt-caching-classivore/

#AnthropicAPI #LLM #PromptCaching #AIEngineering

37% → 78%. Doubled my web scraper's success rate by swapping requests for curl_cffi to mimic Chrome's TLS handshake.

Bonus: deleting 22 lines of "defensive" header overrides added another 2pp. They were undermining the impersonation.

Modern WAFs fingerprint TLS ClientHello and HTTP/2 SETTINGS, not User-Agents.

https://mikenoe.com/posts/tls-fingerprinting-curl-cffi/

#python #webscraping #ai

@caseyliss@mastodon.social how did you learn that your park bench had MMW coverage? Does the phone show an indicator, did you see it on a map, or luck?

1/2 After years as a CTO, I'm building something of my own.

A few months in, the "something" is Classivore — an API that does IAB content classification, brand-safety scoring, and sentiment behind one consistent shape. Predictable per-request pricing, open-source training pipeline on GitHub. If you've ever tried to bolt off-the-shelf classification onto an ad-tech stack, you know how rare both of those are.

#AdTech #AI #Consulting

RE: https://hachyderm.io/@estherschindler/116382916321976268

So it was engagement, not the child porn or racism. OK