AI can do many things extremely well. Something that can do relatively well is to write a university paper.

This is the finding of EduRef, a resource for students and teachers, who conducted an experiment to determine if a deep learning language prediction model known as GPT-3 could deliver successful results in an anonymous trial.

"We hired a team of teachers to create some writing prompts (topics around which you start writing ideas) that were given to a group of graduate and undergraduate students. Also, the same writing prompts were introduced in GPT-3 which did the corresponding research and created the university work, ”states an EduRef publication. The results were an amazing demonstration of AI ability.

This particular AI - GPT-3 (Generative Pre-trainer Transformer 3) - was released in June 2020 by OpenAI, a research firm founded by Elon Musk. It was developed to create content with a human language structure better than any of its predecessors. THE natural language processing has grown rapidly over the last two years, allowing computers to create text.

However, the obstacles to advanced natural language processing are many. According to a 2019 document from the Allen Institute of Artificial Intelligence, the machines essentially have no common sense - the ability to understand what do they write. This finding is based on a critical reassessment of standard tests for determining reasoning logic in machines - such as Winograd Schema Challenge.

This makes the results of the EduRef experiment much more impressive. The writing prompts covered a variety of topics, such as their history USA, research methods (effectiveness vaccine Covid-19), Creative Writing and the laws. The GPT-3 managed to grade the four courses with an average of "C". AI scored the highest on US history and law writing prompts (B- rating in both jobs). The GPT-3 scored a "C" in a research paper on the effectiveness of the vaccine for Covid-19, with a score better than that received by the man-author.

Overall, teacher evaluations showed that GPT-3 was able to mimic human writing in sectors of grammar, the syntax and frequency of the words, although the tasks seemed somewhat technical. As you would expect, the time it took the AI ​​to complete the work was dramatic smaller by that of the participants. The average time between assignment and completion for humans was 3 days, while the average time between it assignment and integration for GPT-3 was between 3 and 20 minutes.

Source: zdnet.com, en.secnews.gr