Has anyone else observed the ChatGPT 3.5 Turbo (0301), accessed via the Azure OpenAI API, generating actual browser links (NOT hallucinations)? This occurrence was unexpected and seemed initially like an anomaly. I’m currently seeking documentation or technical explanations for this behavior but have found none…
The gpt-3.5-turbo-0301 doesn’t have as much anti-hallucination prohibition.
Therefore, it can give web links from its knowledge, and equally can give long-expired links or complete fabrications that look like links.
1 Like