ChatGPT has been gaining a lot of attention in the tech industry since its development by the company OpenAI. The Artificial Intelligence chatbot is used for summarizing complex topics, as well as engage in long conversations. This has caused many other AI companies to rush to make their own large language models (LLMs) – the technology to develop chatbots – and integrate them into other products.
One way to test the effectiveness of ChatGPT was to challenge it with Wordle, a word game from the New York Times. With the latest model known as ChatGPT-4, the performance of the chatbot in these puzzles wasn’t too good. Word games should have been an easy and obvious task for GPT-4 to finish, and most people naturally assumed it would. However, the chatbot was not able to find the correct words.
At the core of ChatGPT is the deep neural network, a complex system that maps inputs to outputs. ChatGPT-4 works with words, and this requires them to be “translated” into numbers by the “tokenizer” – a computer program that maintains a huge list of words and letter combinations as identifiers.
The problem with this is that ChatGPT doesn’t have access to the word’s text, making it unable to realistically reason about the letters. This explains why it struggles with word puzzles, as it doesn’t know the structure of the letters within the words. It made better performance in relation to the first letters of words, but it only provided one valid solution from a six-try opportunity, heavily impacting its success rate.
However, ChatGPT demonstrates excellence in its ability to generate source code. It is possible for future LLMs to use external tools to help it solve certain tasks where it currently struggles, such as the Wordle puzzles.
OpenAI is a research laboratory based in San Jose, California led by globally renowned scientists, engineers and thinkers from a wide range of background. They developed technologies that have and will continue to drive economic impact, environmental solutions and social well being.
Michael G. Madden is the established professor of Computer Science at University of Galway. He specializes in circuit design, low power design, cryptology and algorithm optimization. He has won numerous awards for his research in these fields and is looked up to as a mentor by many aspiring computer scientists.
