ChatGPT, developed by OpenAI, has taken the globe by storm with its ability to summarise complex topics or to engage in long conversations.
However, its latest version, ChatGPT-4, may not be as good at word games as one might expect. Michael G. Madden, an established professor and head of the School of Computer Science at the University of Galway, tested the chatbot on Wordle, a popular word game from the New York Times, and found its performance to be unexpectedly subpar.
Despite being trained on about 500 billion words, ChatGPT-4 struggled with simple puzzles, and its responses were mostly hit-and-miss.
Madden also found that the chatbot was bad at producing palindromes, giving obscure results when asked to do so. The reason for this is that all text inputs are encoded as numbers for AI to process, which makes it challenging for the chatbot to excel at tasks like word games.
Madden suggests that future large language models could address these issues in two ways. Firstly, the training data for ChatGPT-4 could be expanded to include mappings of every letter position inside each word in its vocabulary.
Secondly, future LLMs could employ external tools to complete tasks where they typically struggle, an approach known as Toolformer, which was recently presented in a study.
Read more: Bard vs ChatGPT: The AI Battle for Search Engine Supremacy
Originally published at lahoreherald.com