Hot off the photocopier
Ship log
6 entries & countingFrom prototype to proposal
The interview with Company A changed my picture of the project more than any benchmark had. I'd assumed the assistant would answer questions from the company's customer content. It wouldn't. It was meant to help the people who use Company A's software
Take the full dose →Swapping the database without losing the plot
The whole story →This looks like a question we've answered before
Every FAQ has a handful of questions that make up most of the traffic. Company A estimated that abo…
Read on →Even the yardstick needed calibrating
Rule zero: measure the noise first Before trusting any improvement, I ran the identical setup twice. Same code, same questions, s…
Click for mo’ →Psychometrics for AI
How to test an assistant on documents you've never read, and why the test needed testing too.
Start reading →The bigger model lost
How I chose, benchmarked and re-chose the language model for a local AI FAQ, and why the 15-billion-parameter model got outread by one that does its thinking with 4.
More of this →