How Digby double-checks answers
Updated · 5 min read
Every AI model is sometimes confidently wrong. Digby’s answer is a process: plan the question, ground it in sources, and have a different model check the answers that matter.
01
Why AI gets things wrong
Language models predict plausible text. Most of the time plausible and true coincide; sometimes they don’t, and the model sounds just as sure. Facts that changed recently, numbers, names and anything contested are where errors cluster.
02
Plan the question
Before answering, Digby decides what kind of question it is: a quick lookup, an ordinary question, something contested or high-stakes, or a research request. That decides how much work the answer gets.
03
Ground it
When facts are time-sensitive, disputed or beyond what a model knows, Digby searches the live web and cites what it found. The sources appear under the answer so you can open them.
04
Check it
On contested or high-stakes questions, an independent model — a different one from the model that answered — reviews the answer behind the scenes. If it finds a real problem, the answer is updated and you can see what changed.
05
Go deeper on request
Tap “Dig deeper” under any answer to escalate it to an independent check, then to a full research council of models.
Frequently asked questions
Does every answer get a second check?
No — quick, low-stakes questions get a fast answer. Contested or high-stakes ones are checked, and you can ask for a check on any answer with “Dig deeper”.
Can Digby still be wrong?
Yes. Checking reduces errors but doesn’t eliminate them, which is why Digby shows its sources.
Try asking Digby
Keep reading
Try Digby now.
Free, no account needed. Models and data stay in Canada.