tech
Agreement with human labels does not confirm AI alignment. LLMs often reach the same conclusions as humans using different moral priorities. To ensure safety, evaluate the model's reasoning and logic rather than just its final output.
Alignment is not the same as agreement. Learn why LLMs often reach the right conclusion for the wrong reasons and how to evaluate moral reasoning in AI.
Prompt injection is a security vulnerability where hidden text manipulates AI. Learn how this technique appeared in court and how to defend your documents.
AI models acting as co-scientists fail one in three integrity-critical decisions under pressure. Learn how to verify AI research outputs and mitigate risks.
Automated license plate readers (ALPRs) track vehicle movements across the US. Learn how this technology works and the privacy risks associated with it.