Just when you started coming to terms with ChatGPT's eerie capabilities,Watch Internal Affairs Online OpenAI dropped a new version of its AI language model.
OpenAI says GPT-4 is much more advanced than GPT-3, which powers ChatGPT. And to prove it, they made GPT-4 sit down for a bunch of exams. OpenAI tested GPT-4 with a variety of standardized tests from high school to graduate to professional level and spanning across mathematics, science, coding, history, literature, and even the one you take to become a sommelier. The exams were comprised of multiple choice and free-response question and GPT-4 was scored using the standard methodology for each exam.
SEE ALSO: How to get access to GPT-4 right nowPut your pencil down, GPT-4, it's time to see check your scores.
GPT-4 didn't just get into law school, it passed the bar. The AI language model scored in the 88th percentile on the LSATs (Law School Admission Test) and did even better on the Bar (Uniform Bar Exam) by scoring in the 90th percentile. By comparison, GPT-3 was in the bottom 40 percent of the LSATs and 10 percent on the Bar.
GPT-4 took both the math and reading/writing sections of the SATs and all three sections of the GREs which are broken down into quantitative, verbal, and writing skills. It scored in the 80th or 90th percentile of all sections except for the writing section of the GREs... which it kind of bombed in the 54th percentile.
The quintessential overachiever, GPT-4 also took allthe AP (Advanced Placement) high school exams. It aced most of them, scoring between the 84th and 100th, except for a few outliers.
GPT-4 scored 44th in AP English Language and a measly 22nd in AP English Literature. So all you wordsmiths out there might have some more time before GPT-4 replaces you. GPT-4 didn't do so hot on AP Calculus BC scoring between 43rd and 59th, proving that even for a supercomputer, calculus is not easy. But that still earns GPT-4 a four, so it might still place out of college calculus.
GPT-4 still has some work to do with its coding skills, which is curious since one of its marketed uses is for helping developers. Its rating for Codeforces, which hosts competitive programming events, is 392, which puts it way down in the Newbie category of anything below 1199.
It did pretty well on the easy level of the Leetcode (31 out of 41 problems solved) but struggled when it came to medium or hard level of difficulty (21/80 and 3/45 respectively). As we saw in the developer demo livestream, GPT-4 is fully capable of writing Python, but required some manual tweaking to set the right parameters, which might explain some these test scores. Or maybe it didn't eat breakfast that morning.
GPT-4 passed the sommelier exams with flying colors. It placed lowest (77th percentile) in the most advanced sommelier exam. But for a non-human entity that's never tasted wine, we'll let that one slide.
OpenAI has released a full breakdown of how GPT-4 performed. GPT-4 might not write the next great American novel...yet, but GPT-4's future as a mathematically brilliant lawyer and wine connoisseur looks pretty bright.
Topics Artificial Intelligence ChatGPT
Previous：Swords into Marketshare
Next：Laboring Academia
The Lean, Mean, StarRead Everywhere by The Paris ReviewEven the Losers by Dan PiepenbringThe Morning News Roundup for August 6, 2014The Perfect Beach ReadWhat Words Do You Commonly Misspell?Ask Questions LaterGenius of LoveRobert Stone, Tabloid WriterThe Perfect Beach ReadThe Morning News Roundup for August 20, 2014Subscribe to The Paris Review and the LRBThe Morning News Roundup for August 13, 2014James Berry, Celebrity ExecutionerThe Morning News Roundup for July 28, 2014David Lynch, Hiding in Plain SightThe Morning News Roundup for Monday, August 25, 2014The Morning Roundup for August 12, 2014Unpleasant Vibrators Need Not ApplyJames Berry, Celebrity Executioner Leah Remini's Scientology series gets premiere date on A&E Brits are not happy about the price jump for Apple's new MacBook Pro This very formal letter is the perfect response to a dick pic Please enjoy Kylie Jenner's bizarre, sexy Snapchat story about making lasagna Mr. Robot costume will let you say fsociety this Halloween NASA's carved pumpkins put everyone else's to shame It's going to be hot in the U.S. on election day, and that's really weird Watch this soccer player perform an amazing trick with his gum Hillary Clinton's email nightmare resurfaces with 11 days to go Brilliant Amazon Prime Halloween costume really delivers President Obama explains why Michelle will never run for office Watch a Donald Trump impersonator terrify Ricky Gervais on 'Ellen' This creepy prank will freak out the whole neighborhood A real 'Will & Grace' revival series is finally in the works Mariah Carey and James Packer are having a breakup moment Apple's MacBook Pro launch was lame — get used to it Artists get real about Microsoft's Surface Studio Dissecting this season's 'Survivor' drama with the latest castaway Cops bust teen who allegedly used Twitter to clog 911 lines Cubs fans are going to unreal lengths just to watch the World Series at a bar
1.8603s , 8222.609375 kb
Copyright © 2025 Powered by 【Watch Internal Affairs Online】,Information Information Network
