AI is becoming a lot more accurate than I originally thought. By the end of the decade, I sense AI will be able to recite anything on the spot with precision. During Thursday’s lab period in which the class used three different sources for AI to summarize without using a document, I was surprised at the results. The common guess among the class was that the AI would not be able to pull specific details about the articles without the use of a pdf to upload to the LLM. My group, which had a very popular paper (Group 2), were shocked to see the accuracy of the results ChatGPT gave us on the summary of the paper.
Sure, it helped when a document was uploaded to the server, but the majority of the information was provided without the use of a pdf upload. The other groups had some similarities but group 3 had a bit more hallucinations from AI as the source was not as popular. As these algorithms train over the years, I feel like every source will become incorporated into their algorithms, allowing for the use of a simple search without the use of a pdf to read from.
The issue with this is that we can not consider what is “known” by AI. This simply means that whatever we input to AI such as ChatGPT cannot be considered fully accurate at its current state. Until labs like ours on Thursday start to become concise summaries across several papers without the need of a pdf upload, it will be necessary to always double check the paper itself to check for accuracy. Even then, our paper was more accurate with a pdf attached to the AI though it is always good to cross check papers when summarized by AI, pdf or not.
