3 ms·
Telemetry doesn't guarantee that people won't pull numbers out of their asses. There are lies, damned lies, and statistics. Designing good telemetry is hard, i
by simias 5y ago
Telemetry doesn't guarantee that people won't pull numbers out of their asses. There are lies, damned lies, and statistics.
Designing good telemetry is hard, interpreting the data well is harder (and it's tempting to cherry-pick post facto to justify your decisions). Are users not using functionality X because it's useless or because there's some other issue preventing its use? Or maybe users just don't know that it exists?
Crash reports are more of an unmitigated good IMO however, since they can be used to pinpoint the source of crashes and identify regressions.
- sharklazer 5y ago> Designing good telemetry is hard, interpreting the data well is harder The biggest problem facing any use of data is quality of the collected data. For example, if you wanted to train a model but all you had were examples of a particular class... well, you excluded yourself from being able to predict any other class from the git-go. Data collection itself require a deep understanding of statistics and probability. Otherwise you’re just shooting yourself in the foot. But let’s be honest, machines don’t “learn” things, people do. Machines have no teleology they determine for themselves. People are both the problem and solution. Anyone who thinks big data and ML will be able to generate the answers are deluded and have already turned off or lost completely their critical thinking skills. Data collection in the form of telemetry is only a mechanism for finding solutions and an incredibly INVASIVE one at that.