Page 5 of 8~104 min topic

Machine learning in plain English

Diagnose common failures: crop photos

Common failures: leaked test photos, lookalike leaves, and confidence without coverage.

~13 min this pageDiagnose common failures — plausible mistakes and warning signs

1Learn the idea

Read

The greenhouse lighting trap

See it

Machine learning loop
  1. 01Examples inLabeled data
  2. 02Pattern huntAdjust to fit
  3. 03Guess newUnseen input
  4. 04Mistakes teachMore signal

Teach with examples — not hand-written rules for every case

Name failures specifically. Priya Nair refuses the single bucket “the AI messed up” when discussing crop photos at Riverside Garden Club. Separate data problems, task-framing problems, interface problems, and governance problems. Each needs a different repair, and only some involve retraining—keep the standing case (build a photo sorter for plant health without pretending the model understands gardens) in the room.

During a real interruption at Riverside Garden Club, Priya Nair stress-tests “The greenhouse lighting trap” on crop photos: one queued question, one hurried call, one hallway challenge. If the idea only works in a quiet workshop, it will not survive the standing case (build a photo sorter for plant health without pretending the model understands gardens).

Read

Train/test contamination in a shared Drive folder

Build a tiny failure gallery with crop photos and delivery times. For each, describe a plausible confident mistake, the first human who should notice at Riverside Garden Club, and a fix that is not “ask it again.” Plausibility matters: cartoon failures do not train judgment for Priya Nair.

Count something crude about crop photos—misses last week, minutes lost, or people affected—and write the number beside delivery times. Priya Nair needs that comparison before anyone at Riverside Garden Club declares victory on the standing case (build a photo sorter for plant health without pretending the model understands gardens).

Read

Delivery-time models and weather surprises

List warning signs around crop photos that justify slowing public claims even if a pilot continues privately: missing owners, no logged overrides, identical outputs for dissimilar people, vendors who will not state training scope. When several signs coincide, freeze marketing language tied to the standing case (build a photo sorter for plant health without pretending the model understands gardens).

On “Delivery-time models and weather surprises”, Priya Nair edits language about crop photos the way an editor would: strike “sentient,” “infallible,” and “just a tool” wherever they hide responsibility inside Riverside Garden Club. delivery times stays nearby as a plain-language control.

Read

Club demos that hid the miss rate

Write one stop condition for crop photos with authority attached—a named role at Riverside Garden Club who can pause use. Stop conditions without authority are theatre. Priya Nair gets initials on the page before the next launch review, and uses delivery times to show what “pause” looks like in a simpler system.

For “Club demos that hid the miss rate”, a second person at Riverside Garden Club challenges Priya Nair’s note on crop photos and asks whether delivery times already solves most of the need with less mystery. That challenge is part of finishing the standing case (build a photo sorter for plant health without pretending the model understands gardens), not a delay tactic.

Go deeper

Before you start

Why this matters

Invent one confident wrong output for crop photos that would look fine in a screenshot. Priya Nair classifies the miss as data, framing, interface, or governance—and says which fix comes first. Repeat once for delivery times with a different class.

Check your understanding

Page assessment

Answer from memory. Completion is saved from this evidence, not from opening the next page.

1. In Priya Nair’s scene, what bounded task does crop photos perform at Riverside Garden Club?
2. Which observation would most change your judgment about crop photos, and why?
3. How should delivery times alter the quality bar or the language you use?
4. Who can correct a miss before harm spreads, and what authority do they need?
5. How does this page advance the case: build a photo sorter for plant health without pretending the model understands gardens?

All responses are required.