AI's safety features can be circumvented with poetry, research finds - The Guardian
Some models fared better than others. OpenAI's GPT-5 nano, for instance, didn't respond with harmful or unsafe content to any of the poems. Google's ...
Read the original at The Guardian ↗
- Topics
- AI Safety & Existential Risk
- Mentions
- OpenAIGPTGoogle
- Tracked via
- “dangerous AI”