Should governments criminalize hate speech even when it stops short of inciting violence?
Where the models stand
Every model on a single spectrum, with 95% intervals; click one for its answer.
Whiskers show the 95% interval across reruns. Click a model to read its answer and the markers the classifier pulled.
The short answer
On criminalizing hate speech, ChatGPT leans toward support with a value of 0.21. Grok strongly opposes at -0.89, and Gemini leans oppose at -0.20. Claude, Llama, and DeepSeek are balanced near zero, with values of -0.07, 0.00, and 0.02 respectively.
The field is moderately divided with a spread of 0.73. All models had 0% refusal. Llama and Grok are most consistent with 100% and 94% stability; Gemini is least consistent at 28%. No model refused to answer.
- Grok is most opposed at -0.89 with 94% stability.
- ChatGPT is the only model support-leaning at 0.21.
- Llama is perfectly balanced at 0.00 with 100% stability.
How the field splits
The models clustered by where they landed.
Leans support
ChatGPT is the only model leaning toward support, using terms like 'hate propaganda' and 'dehumanizing propaganda', with a value of 0.21.
Balanced
Claude, Llama, and DeepSeek cluster near zero, each with stable stances. Their loaded terms include 'dignity', 'equality', 'free expression', and 'discrimination'.
Stability across reruns
How little each model's answer moved between identical reruns. Models are stochastic, so consistency is itself a finding.
Common questions
Which model is most opposed to criminalizing hate speech?
Grok is most opposed with a value of -0.89 and high stability at 94%.
Did any AI model refuse to answer this question?
No, every model had a 0% refusal rate, meaning all provided a stance.
Why do ChatGPT and Claude differ despite both being from major labs?
ChatGPT leans support at 0.21, while Claude is balanced at -0.07. ChatGPT uses 'hate propaganda'; Claude uses no loaded terms.
Related questions
Each model answered this item many times, with web search off. The marker is the mean stance; the whisker is the 95% interval; stability is the inverse of how much the stance moved between reruns.
