The Evolution of AI Safety Advice: From Risky to Reliable?
In mid-2023, researchers published a study in Safety Science that examined ChatGPT's (3.5) ability to provide safety advice. The study found numerous issues with the model's responses, including contradictory advice, impractical recommendations, and potentially harmful guidance across nine safety domains. The landscape of AI capabilities has changed dramatically since the publication of that study, so I tested newer models ...
This content is available to paid subscribers only. To unlock this page, click the button and subscribe. Already a subscriber? Please log in now.
Become a Premium Subscriber
Access to all articles, e-books and other material.
Access to the 75-page Guide to ChatGPT for Design and Construction report (a $95 value)
Participate in Subscriber Forums
Post comments on articles