Profile avatar
faz.ms
Interested in #Infosec and #ProductiveDisagreement | #StayAtHomeDad, worked in #Biotech, co-built tiny companies in renewables and structural engineering sectors, ex #HumanRights observer | likes #ElixirLang, #Boardgames |🔸 #10PercentPledge
243 posts 150 followers 987 following
Getting Started
Active Commenter

Job security for security consultants has been achieved!

me: mom, can we have scope sensitivity? mom: no. there is scope sensitivity at home. *at home: whataboutism*

„If hallucination matters to you, your programming language has let you down.“

„All models are wrong, but some are useful“ is truer than ever

Actually implementing a system that uses models of different sizes on datasets, and coaxing them towards consistent output, can really feel like summoning some trickster spirit. "You want JSON objects? no problem here are a few extra properties I made up. No? Ok, I'll just rate half of the data."

As a child I never would have guessed the order that we would get Star Trek technologies. Having a computer that you can just tell what you need and modify the result by talking to it more seemed a lot less realistic than the warp drive.

Outline what I want to build, have o3 to turn it into a PRD, discuss it until I get an architecture that feels right. Use taskmaster-ai via MCP in Cursor to turn the PRD into tasks and subtasks, then lead Gemini 2.5 pro through each subtask on a short leash while feeding it dependency docs.

mood

Ouch, getting dunked on by my IDEs AI autocomplete suggestions 😭