Listen to this story
ChatGPT creator OpenAI yesterday released a new update– “with improved factuality and mathematical capabilities.”
So, we tried.
Turns out, it is still making the same errors. We knew than ChatGPT is bad at maths, but we did not know that even after the latest upgrade of Jan 30th, it is still giving incorrect results.
Read more: Why is Everyone Bashing ChatGPT?
Not just maths, it is still proficient in giving factually incorrect answers as it is trained on pre-2021 data.
New Update, Old Joke
Recently, a post has been going around the social media where ChatGPT hilariously agrees to a person when he coaxes it to believe that his spouse is right even though the answer is wrong.
Last month, Stack Overflow blocked all answers from ChatGPT citing inaccurate responses from the bot. Although Stack Overflow stated that this ban was temporary, it emphasized that the issue was not just with the incorrect answers but also with how the bot phrased them.
ChatGPT is based on GPT-3.5, which can generate sentences that are grammatically correct and have a formal tone. However, as per the nature of LLMs, this also makes the responses seem overly confident and authoritative, which can be misleading.
Despite ChatGPT has passing the bar exam and MBA finals, it still has the problem of bias and lacks logical reasoning besides being terrible at maths.
Although OpenAI claims to be working towards minimizing wrong answers, no one knows how long shall it be, especially since it is a dedicated responsible AI research lab.CEO Sam Altman took to twitter to share how the team is working towards making ChatGPT perfect.
Read more: Freaky ChatGPT Fails that Caught Our Eyes