WEBVTT

00:00:00.000 --> 00:00:05.157
A compelling technology announcement is a starting point. To make a decision,

00:00:05.157 --> 00:00:10.313
separate the claim, the measurement, and the conclusion you are considering.

00:00:10.313 --> 00:00:15.028
Start by identifying the exact claim. Faster can mean latency, throughput, or a

00:00:15.028 --> 00:00:19.744
single benchmark score. Name the workload, the comparison, and the conditions

00:00:19.744 --> 00:00:24.459
that make the statement meaningful.

00:00:24.459 --> 00:00:29.562
Next, inspect how the result was obtained. Look for the hardware, software

00:00:29.562 --> 00:00:34.666
version, dataset, repetitions, and failures. Missing conditions do not prove the

00:00:34.666 --> 00:00:39.769
claim is false. They limit what you can conclude.

00:00:39.769 --> 00:00:44.224
Ask whether the test represents your use. A result on one dataset or a short

00:00:44.224 --> 00:00:48.679
burst of activity may not describe sustained performance, accuracy, cost, or

00:00:48.679 --> 00:00:53.134
reliability for your workload.

00:00:53.134 --> 00:00:57.984
Record the demonstrated result separately from your expectations. State the

00:00:57.984 --> 00:01:02.833
assumptions. Then identify the smallest responsible test that could resolve the

00:01:02.833 --> 00:01:07.683
uncertainty without risking important data or systems.

00:01:07.683 --> 00:01:11.847
Now practice with a technology headline. Write one sentence describing the

00:01:11.847 --> 00:01:16.012
claim, one describing its evidence, and one describing what is still unknown.

00:01:16.012 --> 00:01:20.176
The goal is not to sound skeptical. The goal is to know what your decision rests

00:01:20.176 --> 00:01:24.340
on.