~/talk

The Algorithm That Nearly Killed Me: When Testing Isn't Enough

Talkat NDC Oslo 2026 · 2026-09-14..09-18· [machine-learning, testing, tech-ethics]
links:none on file
slides:none on file

An insulin pump system where the control algorithm on the phone confidently reports high insulin delivery while the hardware actually sends zero units causes blood sugar to climb into dangerous ketones and near diabetic ketoacidosis in 72 hours. The app’s learning amplifies bad training data; the device defaults to 1 unit/hour when disconnected and some models dose 0.1 units per hour as a safety drip — small numbers with large effects. Tests show green ticks but the real system fails; validation must check decision boundaries, invariants and overlap of edge cases.

AI-written summary · about

  1. 0:00Introduction and personal story
  2. 10:30Diabetes management and pump transition
  3. 26:58Algorithm failure and edge cases
  4. 39:32Testing responsibility and industry gaps
  5. 41:00Ten strategies for impactful testing
  6. 59:37Conclusion and user safety
~/search
type to search 6,518 conferences, 143,918 speakers and 192,133 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
← →
on a talk: 5 seconds back / forward
0–9
on a talk: jump to 0%–90% of the recording
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.