Here’s Why ChatGPT Isn’t As Great As You Think

(CTN News) – ChatGPT, developed by OpenAI, has become very popular worldwide. Everyone has been caught off guard by its articulation and database. It has been noted that the chatbot is capable of summarizing complex topics or conversing with a large number of individuals at once.

It is not surprising that rival AI firms are scrambling to publish their own large language models (LLMs), which power chatbots like ChatGPT. The LLMs will be included in other products, such as search engines.

Michael G. Madden, Chair of the School of Computer Science at the University of Galway, chose to test the chatbot with Wordle, a New York Times word game.

A five-letter word can be guessed six times by a player. In each guess, the game indicates which letters are in the correct position in the word.

In his analysis of these puzzles, he discovered that ChatGPT-4 performed unexpectedly poorly.

Due to the fact that LLMs are trained in text, one might expect AI at the level of GPT-4 to be an expert at word games. According to reports, GPT-4 was trained on approximately 500 billion words.

It was Madden who tested GPT-4 using a Wordle puzzle that he was familiar with. As a result of ChatGPT-4, five out of six responses were unsuccessful.

While the chatbot was able to find some solutions, its performance was mostly hit and miss.

Madden states that ChatGPT-4 is also unable to handle palindromes. If asked to produce any results, it produces obscure results.

Initially, it may seem surprising that GPT-4 struggles with simple puzzles, but that is due to the fact that all of our text inputs are encoded as numbers for AI to process.

It is possible for future LLM students to get around this in one of two ways, as explained by Madden. Considering that ChatGPT-4 is known to understand the first letter of every phrase in its vocabulary, its training data might be expanded to include mappings of every letter position within each word in its vocabulary.

Compared to the first option, the second is more comprehensive and interesting. It is possible that future LLMs will produce code that addresses issues such as these.

A study presented an approach known as Toolformer, where an LLM uses external tools to complete tasks where they normally struggle, including mathematical operations.

SEE ALSO:

Lululemon Sees Sales And Profits Above Estimates In 2023