Sentiment under stress

Try praise, criticism, negation and sarcasm. Watch a real classifier assign its scores.

How this experiment works

A quantised DistilBERT model fine-tuned on SST-2 produces a probability distribution over two labels.

What it does not prove

This English movie-review model predicts positive or negative sentiment. It cannot assess truth, safety, intentions or mental health.

Try this challenge

Find a sentence whose meaning changes when you add “not”. Does the model follow?

Related Trust Agent courses · Sources and licences

Enable JavaScript to run the experiment on your device.