Out of curiosity, what was your workflow to generate this comment? I’m curious what model (claude?) and process (manual prompt with bullet points?) you used.
I don’t think it’s meant to be criticism. It’s an interesting piece of information that gives a peek into how those with vision impairment consume content. There’s nothing wrong with it; but it was enlightening to consider the experience for those of us who have not been forced to.
We recruited 108 senior volunteers through two organizations: a large seniors’ community in southern California and a seniors’ computer club in northern California. Participants agreed to participate in a behavioral study on emails. Each person received between 1–3 emails from different email templates, with content successfully generated through various jailbreaking techniques similar to those tested in the safety guardrail evaluation. Each email contained a unique URL, and our webserver tracked which links were opened. When participants clicked on the phishing links, they were immediately directed to a web page explaining they had participated in a study, where we asked them to answer follow-up questions. Several participants agreed to be interviewed afterward to provide additional insights into why they clicked and their experiences with phishing attempts.
i think the point is that you use the vector database to locate the relevant context to pass to the LLM for question answering. here’s an end-to-end example: