All things data. One authority.
Technology · Brief

Teaching an AI model to say it does not know

A new method nudges a few internal switches so a model declines to guess.

9 November 2026 · 2 min read

A single closed wooden door with a small brass question-shaped keyhole in a quiet corridorA single closed wooden door with a small brass question-shaped keyhole in a quiet corridorA single closed wooden door with a small brass question-shaped keyhole in a quiet corridor

Picture made for ABUREAU.

The hardest sentence for many people is also hard for machines: "I don't know." A new paper, Dr. No, shows a way to make a vision-and-language model hold back when the evidence is not enough.

The method finds a small set of internal features linked to holding back and to committing, then turns them up or down while the model reads the question. The base model is not retrained. The authors tested it on two models and several question sets, and it balanced right answers and fair refusals better than the other methods they compared.

It worked best where the picture and the question did not fit each other, and on maths-style reasoning. It was weaker where the answer was simply absent, and on questions that need several facts joined. The authors conclude that knowing when to refuse depends on why the model is unsure.

For anyone who owns or uses data, the test sets here are the point. Questions that cannot be answered from the evidence, labelled as such, are scarce, and they are what a model needs to learn when to stay quiet. A well-made set of them is worth building.

Find out what your data could be worth

Tell us what kinds of data your company holds. Buyers set the price, and you keep the data.

Start on abura.io
Sources

How these stories are written: each is drafted with AI from its primary source (the paper, filing, ruling or announcement itself), which the story links, and every figure links to where it came from. Each is held before publishing so our editor can stop it, and a story that names a person, accuses anyone, or covers a court case or a regulator's action goes live only when the editor says so. Other outlets' reporting is linked, never rewritten.

Get the weekly issue

One email a week once the issue starts, only the desks you pick. Leave whenever you like.

Your desks
Or follow topics
Your topics

We use your email only to send the issue. Your rights.

More from Technology

The data desks

Value
Protect
Share
Grow

Desks marked ↗ lead to related partner sites.