AI experts recognize the fact that neural networks are often wrong. On this occasion, they are not part of the Russian Federation, for example, to write a pragma code. However, there was no final point of view on this issue. To gain clarity, the ChatGPT chatbot questions were asked to answer a large number of questions related to programming. The neural network is focused on correctly answering only a small part of them.
The test was conducted by researchers at Purdue University and identified over 500 questions that raise highly specialized programming problems. The results obtained were submitted for examination to expert volunteers. At the same time, not only the correctness and consistency of the answers were evaluated, but also their conciseness and completeness. In addition, the team analyzed the linguistic patterns and overall mood of each of the responses. As it turned out, only 48% of AI questions from OpenAI often give a sane answer.
Users found 40% of the answers “believable” to themselves, although 77% of that number were wrong. The human experts identified the error only when it was obvious. If errors were identified when identifying the middle stage of development, studying the documentation, then users underestimated the likelihood of errors in AI responses. At the same time, the pleasant, confident and positive tone of presentation is partly to blame. The completeness of the answers and the use of a simple intelligible style were met in such a way that even incorrect answers were replaced by real ones in the observations. inspectors.
Source: Tech Cult
I am a professional journalist and content creator with extensive experience writing for news websites. I currently work as an author at Gadget Onus, where I specialize in covering hot news topics. My written pieces have been published on some of the biggest media outlets around the world, including The Guardian and BBC News.
