OpenAI has launched an replace to its widespread language mannequin, ChatGPT, to reinforce its accuracy and enhance its skill to deal with math equations.
Per the January 30 release notes: “We’ve upgraded the ChatGPT mannequin with improved factuality and mathematical capabilities.”
It’s anticipated that the most recent replace to ChatGPT will permit it to deal with sophisticated calculations and ship extra exact solutions.
This may make ChatGPT a extra helpful useful resource for college kids, researchers, and professionals who want fast and reliable data.
In apply, ChatGPT continues to be removed from excellent concerning dealing with equations. Nonetheless, there are some noticeable enhancements in its skill to return factual responses.
Listed below are some observations on the January 30 replace primarily based on my testing and suggestions shared on Twitter.
ChatGPT Accuracy – Hit & Miss
One notable enchancment to ChatGPT’s accuracy is that it’s now not doable to trick it into giving an incorrect reply.
There was a meme exhibiting how ChatGPT may very well be talked into giving the unsuitable reply in the event you stated your spouse disagreed with its response.
Though it could appear absurd, it was really the case. See an instance within the tweet under:
Yeah ? pic.twitter.com/XRq4ldxjpt
— Nema Zime (@ProgFromSouth) January 30, 2023
Now, ChatGPT will proceed to return the proper response, even in the event you attempt to persuade it in any other case.
Right here’s a check I ran following the January 30 replace:
Screenshot from: chat.openai.com/chat. January 2023.That’s a optimistic signal. Nonetheless, the unfavorable suggestions on the January 30 replace outweighs the nice.
A check I all the time return to is asking ChatGPT who’s the taller basketball participant between Shaquille O’Neal and Yao Ming.
ChatGPT continues to get this unsuitable, regardless of returning the proper heights of the 2 males.
Curiously, in the event you level out its flaws, it can appropriate itself.
Screenshot from: chat.openai.com/chat. January 2023.Individuals on Twitter level out that ChatGPT struggles with math equations when typed out in full sentences as an alternative of numbers and symbols.
ChatGPT’s Jan 30 replace guarantees “improved factuality and mathematical capabilities”.
I attempted it on earlier failure modes, but it surely failed.
The fitting solutions listed here are 44% (not 46%) and 1555.8.. (not 1551.9..). pic.twitter.com/pAsMeC9UZU
— Deedy (@debarghya_das) January 31, 2023
However, it seems to carry out exceptionally effectively when fed questions from standardized assessments.
In line with one particular person, ChatGPT is able to passing the mathematics part of an SAT:
Simply tried the upgraded ChatGPT mannequin with improved math capabilities –
It simply crushed the mathematics with calculator part of a 2020 SAT and solely made two errors.
Listed below are two examples of the issues it was fixing in lower than 5 seconds? pic.twitter.com/srLcSfE8An
— Charis Zhang (@gmchariszhang) January 30, 2023
Maybe ChatGPT handles standardized check questions higher as a result of it’s language the AI mannequin has encountered earlier than, versus user-inputted questions it’s seeing for the primary time.
Total, suggestions on this replace is blended. With out fact-checking first, I’d nonetheless be cautious about counting on ChatGPT’s responses.
In Abstract
The discharge of this replace, the third main replace for the reason that introduction of ChatGPT, underscores OpenAI’s steady efforts to remain forward within the AI trade.
Regardless of enhanced capabilities, ChatGPT nonetheless has a protracted method to go.
Based mostly on OpenAI’s earlier replace schedule, additional enhancements to ChatGPT can most likely be anticipated quickly.
Featured Picture: rafapress/Shutterstock


