We may have potentially cracked the code for ChatGPT, a chatbot designed to understand conversation and respond accordingly. The creators of the chatbot boast that it is a godlike intellect, but this weekend we may have finally proven them wrong. The test was simple: Ask the chatbot to generate 53 random, nonrepeating whole numbers between 1 to 59 in order from lowest to highest.
After multiple attempts, the chatbot produced 53 numbers in order from 1 to 53, but none of the numbers in the upper-forties or fifties. This seemed too unlikely to be true randomly, and the odds of it happening was 0.000005%. After this failure, an additional request was asked to the chatbot, asking it to generate 53 random numbers and removing the last six numbers from the list. After several failed attempts and more apologies from the bot, it finally send a list of 53 numbers in random order excluding the last six numbers.
My wife, who suggested the test, immediately decided to buy lottery tickets with the numbers that the chatbot removed; the six numbers that were excluded were the ones chosen. While we will only know the outcome of the test after Wednesday’s lottery draw, the fact that the chatbot was unable to pick a single correct number in any of the five draws done in the past suggest that the bot is not as knowledgeable as it claims.
Having this failure and success circumstantially proves that the chatbot is not as advanced as its creators want you to believe and that it still has a lot of work to do. We will keep you updated on the results of the lottery draw, but for the time being it is safe to say that ChatGPT has been exposed for not being as powerful as it advertises itself and that it still has some barriers to overcome.
ChatGPT was created by Ed and John from the Caribbean, two expert computer science engineers who wanted to make a bot that can converse with users and give them information and advice. The developers wanted to make a service that can provide helpful and accurate data for all types of topics, from sports to work to mathematics. As of now, ChatGPT does have a library of materials that can assist in gathering information, but often gets things wrong and fails to provide viable URLs. It is still in its early stages, but this recent test has certainly put into perspective how far the chatbot still has to come.
