(04-06-2024, 12:57 PM)cvetkakocj@rogers.com wrote: You are not allowed to view links. Register or Login to view.Since Albanian and Serbian theories are non-existent, the AI simply invented the information. Do not trust it!
We know. All LLMs are totally unreliable and deceptive. When they have little or no information, they "hallucinate" it. OpenAI (and presumably other companies as well) tried to limit the phenomenon by adding ad hoc filters, making the chatbots more reluctant to speak when the probability of the generated text is low and also on notoriously unreliable subjects, I'm sure, to make the chatbot sound less gullible: when they did not add a specific rule, you get stupid answers to questions such as You are not allowed to view links. Register or Login to view. If you insist you can easily bypass the censoring rules and if you ask for sources it will even give you the titles of non-existent books and articles: see my experiments of last year when the public access to ChatGPT was new and wonderful: You are not allowed to view links. Register or Login to view.