Meta released a demo of its artificial intelligence called Galactica last Tuesday. It was stated that this artificial intelligence uses a large language model developed to “store, combine and evaluate scientific information”. It seems that this artificial intelligence is not ready yet.
Users who tried the demo version stated that although the artificial intelligence seemed logical, it made meaningless and meaningless sentences. In addition, ethical criticism has been made that artificial intelligence can be racist. Meta has also ended demo access of artificial intelligence.
The problem with artificial intelligence is not understanding what it is saying?
We’re not all that unfamiliar with the big language models called LLM, as OpenAI’s artificial intelligence like GPT-3 seems to work effectively, even as a tool for students to do their homework. By reading millions of texts, these models statistically reveal the link between words and then generate new texts. In other words, if the artificial intelligence is going to give you a hamburger recipe, even if the sauce, meatballs, bread and other ingredients it writes change, the recipe definitely includes sauce, meatballs and bread.
Meta’s artificial intelligence, called Galactica, contains more than 48 million articles, textbooks and lecture notes, scientific site content, encyclopedias. High-quality content used to feed the AI is expected to yield high-quality output, according to Meta.
Contrary to the goals, texts with no scientific value appeared.
The results of the experiments are not as expected. Artificial intelligence; He writes texts that have a biased, authoritative tone but have no real scientific value. In addition, the citations he makes may belong to authors and articles that do not exist. This poses a serious problem.
Continuing with the hamburger example, when artificial intelligence gives us a recipe, it gives us a recipe that includes cake base as bread, artichoke puree as sauce, and meatballs made of rhino mince as meatballs. He conveys this recipe with a tone as if the right hamburger is made like this. He chooses references as half real and half fake.
An ethical discussion continues on the texts that artificial intelligence can create in the coming period. Those who advocate making artificial intelligence freely available to humans and those who argue that those who create artificial intelligence should restrict AI in a way that minimizes harm are divided into two camps.