Can you trust an AI chatbot? There are a lot of examples of AI chatbots generating erroneous data – so maybe not. This article in The Messenger Tech reports Microsoft’s own research efforts in the ‘trustworthiness gaps’ in GPT-3 and GPT-4, both used by Windows Copilot:

The research focused on OpenAI’s GPT-3 and GPT-4 language models and evaluated their levels of toxicity, bias and stereotyping, robustness to adversarial attacks, privacy, ethics and fairness. 

Ultimately, the researchers found that “the trustworthiness of GPT models remains limited.” They also discovered that the GPT models have a tendency to generalize when asked about ongoing events outside their scope of knowledge.

What does this mean? It means be very, very careful before acting on a ‘fact’ generated by an AI chatbot. When in doubt, check with a human.

Read the article here.