An expert explained the mysterious paradox of ChatGPT
Share Source: The Verge
Many users of large language models have long noticed that even the most advanced AI assistants sometimes cannot answer the simplest questions. For example, ChatGPT still cannot tell the current time.
Main theses:
- The problems with AI are not limited to determining the current time.
- He also has a poor understanding of any tasks involving the real passage of time.
ChatGPT paradoxes shock users
The key caveat is that ChatGPT does not have access to a clock, so it often tries to guess the time.
Robotics expert Yervan Kulbashyan points out that the reason lies in the structure of the generative models themselves.
According to him, they don't “know” anything in the literal sense of the word, they just predict the most likely answer based on training data.
The expert also explained that this can be compared to a situation where you have a huge library but no clock.
Technically, companies can give AI access to the current time, and in cases where web search is enabled, ChatGPT is indeed able to answer correctly. But this creates a side effect: the models have a limit on the amount of information they can hold at once.
Share
Yervan Kulbashyan also emphasized that if the current time is constantly loaded, it will take up part of this space and interfere with the processing of more important data.