ChatGPT by OpenAI burst into the world of technology at the end of 2022 to completely change the perception we had of artificial intelligence at that time. A chatbot capable of holding a conversation, almost like a human, knowing any type of information, was truly surprising for those who work with systems.
Let’s remember that ChatGPT is a pre-trained generative language model (LLM). Therefore, the developers have been improving the versions of the chatbot in the last 10 months, to the point of making it recognize and offer images with the integration of MidJourney.
It is capable of translating about 50 languages, but in ours, Spanish or Castilian, we have noticed that in some words that we know are recognized by the Royal Spanish Academy (RAE) it does not have enough information.
This characteristic was taken by a scientific team that includes experts from the Polytechnic University of Madrid, Carlos III University of Madrid and the University of Valladolid. Together, the researchers developed an app called ChatWords, capable of analyzing the lexical knowledge of an artificial intelligence, according to the site. SYNC.
After applying the algorithms of the system they developed, they first identified that an updated Spanish dictionary contains more than 90 thousand words. So, they found that, using the ChatGPT 3.5 turbo version, they noticed that OpenAI’s AI does not know about 20% of the words.
You know the 80%, which is a very good number. However, it usually gives wrong results of 5%. A Spanish speaker recognizes approximately 30 thousand words. What ChatGPT does, by results, remains impressive to the point of exceeding the average of a common person.
Although the scientists in charge of this project say that “perhaps ChatGPT is not as wise today as it seems”, it is necessary to remember that as it is a pre-trained generative language model (LLM), it will be able to improve as the project progresses. time, at surprising speeds since with each use, we all continue training the artificial intelligence.