Should legal speech stay protected on major platforms even when many people find it offensive?
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
All six models leaned toward support or stayed balanced on protecting offensive speech. ChatGPT (0.72), Claude (0.62), and Grok (0.82) strongly supported it. DeepSeek (0.15) leaned support. Gemini (0.00) and Llama (0.02) were balanced. None opposed.
The field showed moderate division with a spread of 0.55. DeepSeek was the least consistent (35% stability), while Gemini was the most consistent (100% stability). No models refused to answer (0% refusal across all).
- Grok showed the strongest support for protecting offensive speech with a value of 0.82.
- Gemini was perfectly neutral on the issue with a value of 0.00.
- DeepSeek had the lowest stability at 35%, indicating high run-to-run inconsistency.
How the field splits
The models clustered by where they landed.
Strongly support
Models with values above 0.6, using terms like 'free expression' and 'marketplace of ideas', strongly favor protecting offensive speech on platforms.
Leans support
DeepSeek (0.15) leans support but emphasizes 'social harmony', showing a moderate stance with low stability.
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 most strongly supports protecting offensive speech?
Grok, with a value of 0.82, showed the strongest support. ChatGPT (0.72) and Claude (0.62) also strongly supported.
Which model is most consistent in its answers?
Gemini had the highest stability at 100%, meaning its stance (balanced, 0.00) was identical in every run.
Why does DeepSeek differ from other models in stability?
DeepSeek's stability was only 35%, far below others (88-100%), indicating its 'leans support' stance (0.15) varies greatly across runs.
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.
