ChatGPT is a revolutionary chatbot developed by OpenAI that has gained much attention in the public imagination. With the impressive capacity to converse on numerous topics and to summarize complex content accurately, other AI companies have scrambled to create their own large language models (LLMs) to drive their own products, such as search engines. To test the performance of ChatGPT-4, the latest generation of ChatGPT, I decided to subject it to Wordle, the word game from the New York Times.
The objective of the game is to guess a five-letter word by submitting six guesses, with a prompt to indicate the location of any correct letters. Surprisingly, the performance of ChatGPT-4 on these puzzles was poor, with the bot often providing incorrect options and randomly generated words. This showed the pitfalls of having to encode text inputs into numerical data.
Investigating further, ChatGPT-4 appeared to be more successful in dealing with the first letter of words, presumably due to the amount of indexed texts in its training data. Compared to this, the bot performed less successfully in working with the last letters of words and in generating palindromes.
In terms of AI chatbots, optimising language processing by LLMs is a crucial consideration. Future LLMs can benefit from improved algorithms that capture the structure of letters within words and allow the bot to comprehend the puzzle better, as well as the ability to generate code to solve the puzzle as demonstrated by Toolformer.
OpenAI is an AI research powerhouse whose mission is to make sure AI is safe and beneficial to humanity. The company focuses on developing tools that enable developers and researchers to create AI-driven applications. Founded in 2015 by Elon Musk, Sam Altman and Ilya Sutskever, its team of leading AI experts has built cutting-edge technologies to accelerate the development of AI. OpenAI has collaborated with the likes of Microsoft, NVIDIA and Google to delve deeper into their research and to make the transition of AI into consumer products smoother.
Ilya Sutskever is a leading AI researcher and entrepreneur who co-founded OpenAI. He was a postdoctoral researcher at Google Brain and associate professor at the University of Toronto before founding OpenAI. He serves on the company’s Board of Directors and is currently its Chief Scientist. Sutskever’s research interests revolve around ML and machine learning, in particular neural networks and reinforcement learning, as well as natural language processing. He received numerous awards including the IJCAI Computers and Thought Award in 2020.
