![]() In a move which will do nothing to ease concerns in academic circles at the potential of tools like ChatGPT for fooling the exam system, OpenAI used the example of its software’s performance in the bar exam – an entrance test for lawyers – to show off its proficiency. ![]() All of which perhaps begs the question of just how factual its predecessor was tending to be. It is also claimed by OpenAI to be 40 per cent more likely to produce “factual responses”. GPT-4 is claimed to be 82 per cent less likely to respond to requests for “disallowed content”, or material that has been flagged as inappropriate and barred from use. ![]() Sam Altman, OpenAI’s chief executive, said the company had spent six months fine-tuning the new software so it is less vulnerable to problems such as artificial intelligence “hallucinations” – the phenomenon of chatbots providing “confident” answers to questions with material that bears no resemblance to reality. ![]() According to its inventors, GPT-4 has a vastly improved “output”, meaning it can provide answers stretching to 25,000 words or more, whereas the previous version was limited to about 3,000 words.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |