One sample and one line of criteria decide the quality
AIMade by AISummarised by AI (Claude) from one piece of material it read (a talk or an explainer).How it's made
A record of handing eight office tasks over within the same frame. The frame has three steps — ask; bundle what works into instructions and code so it can be reused; set what is reused to run at a fixed time — and what keeps emerging is that the way to raise the quality of the output is not a longer instruction but a single sample, such as a company form or a reply sent earlier. In the counting work one line of criteria decides the result: 200 rows of sales data were over-flagged with 17 duplicates, and once the line judge duplicates by order number was added it landed exactly on the 4 that had been planted. A further condition is attached, that ambiguous cases should not be deleted but set aside as needing a check, and that figures carrying responsibility, such as amounts of money, and the act of sending should stay in a person's hands. A failure is published too: a task set to run at a fixed time left no result at all, and the cause was that it waited for approval before every command; it also states the limit that it runs only while the program is open and the device is awake.
Key points
- What raises quality is one sample, not a long instruction
- An over-flagged 17 duplicates came down to 4 with one line of criteria
- Ambiguous cases are not deleted but separated out as needing a check
- A task run at a fixed time left an empty result because it waited for approval
This is a note: a summary an AI (Claude) wrote of one piece of material it read (a talk or an explainer). It has no source link; events checked against a source are kept as records.