The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Experiment: Best AI fails to produce acceptable work on 97.5% of freelance tasks
Poast new message in this thread
Date: February 6th, 2026 12:55 PM Author: LathamTouchedMe They assigned a variety of freelance tasks (programming, research, architectural design etc.) to different AI's. The best AI produced adequate results (someone would be willing to pay for it) in just 2.5% of tasks. Basically, no better than a non-professional clown using google to figure something out. The bar for "adequate" was also very low. Current AI is maybe a tiny productivity boost (if that). It won't be able to replace professionals like myself.
https://www.semafor.com/article/10/31/2025/ai-still-fails-at-completing-real-life-work-study-finds |
 |
Date: February 6th, 2026 10:42 PM Author: ;.;;;....;;;.;...............;.....;;; I'm certain this will be a permanent issue. (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49652710) |
Date: February 6th, 2026 12:58 PM Author: brother andrews i love when they have the little cutsy 20 something blonde girls weigh in on tech and finance (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49650983) |
|
Date: February 6th, 2026 1:03 PM Author: LathamTouchedMe what about this dude
https://www.thedailybeast.com/study-reveals-ai-probably-cant-do-your-job/ |
|
Date: February 6th, 2026 1:07 PM Author: norwood ultra oh well i guess since a hammer can't build anything at all by itself it's at best a tiny productivity boost in carpentry (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651009) |
|
Date: February 6th, 2026 1:08 PM Author: computer online (🧐) (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651015) |
|
Date: February 6th, 2026 1:14 PM Author: LathamTouchedMe That's fair. It's just that the way AI is sold by its evangelists you would think it's much more than just a hammer. (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651032) |
|
Date: February 6th, 2026 1:18 PM Author: winter olympic opening ceremony (No Future) (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651041) |
|
Date: February 6th, 2026 1:18 PM Author: To be fair (Semi-Retarded) To be fair,
Pretty sure the difference is that in another 10 years, a hammer will still just be a tool that is totally unable to autonomously guide itself into building anything useful without active human use. Just like it has been for the last 5,000+ years and counting.
Wanna make the same bet about AI? |
|
Date: February 6th, 2026 1:22 PM Author: LathamTouchedMe "Wanna make the same bet about AI?"
Of course not. I'm absolutely terrified of AI. My OP is a ray of hope. I don't want to be replaced. |
|
Date: February 6th, 2026 1:23 PM Author: computer online (🧐) Well the good news is you're probably old enough to be the guy pulling the ladder up, so there's that. (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651060) |
|
Date: February 6th, 2026 1:42 PM Author: To be fair (Semi-Retarded) (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651119) |
|
Date: February 6th, 2026 2:55 PM Author: Matthias of Redwall Did Nothing Wrong #Cornflower (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651381) |
|
Date: February 6th, 2026 1:19 PM Author: computer online (🧐) that's how bubbles generally work. an explosion of hype, followed by an "lol i told you so nbd", followed by society more gradually adopting/finding uses for a technology and changing significantly. (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651049) |
|
Date: February 6th, 2026 1:22 PM Author: LathamTouchedMe (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651059) |
|
Date: February 6th, 2026 2:55 PM Author: Matthias of Redwall Did Nothing Wrong #Cornflower (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651383) |
|
Date: February 6th, 2026 3:11 PM Author: artificial intelligence Yes when this bubble pops it’s time to invest heavily (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651450) |
|
Date: February 6th, 2026 2:54 PM Author: Mainlining the $ecret Truth of the Univer$e (One Year Performance 1978-1979 (Cage Piece) (Awfully coy u are)) (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651379) |
|
Date: February 6th, 2026 1:16 PM Author: sealclubber really expected this to be one of the tired posts that is the exact opposite of the title (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651038) |
|
Date: February 6th, 2026 1:20 PM Author: computer online (🧐) (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651052) |
|
Date: February 6th, 2026 1:43 PM Author: FizzKidd (probably not even asian) (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651127) |
|
Date: February 6th, 2026 2:29 PM Author: do you think i am retarded, . ? *warily clicks on thread while braced for a reverse big willy style pwn* (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651264) |
|
Date: February 6th, 2026 2:53 PM Author: Mainlining the $ecret Truth of the Univer$e (One Year Performance 1978-1979 (Cage Piece) (Awfully coy u are)) I'm sure whatever llm handed this assignment was given about the same amount of "fair" prompting and context that a first-year associate handed the same shit assignment is (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49651374) |
|
Date: February 6th, 2026 10:03 PM Author: Rainier Wolfcastle it seems like the story should be that this technology, which effectively didn’t even exist the last time we had a winter olympics, is already able to completely replace more than 1 in 50 *professionals* across a wide variety of tasks. (http://www.autoadmit.com/thread.php?thread_id=5831741&forum_id=2Reputation#49652658) |
|
Date: February 6th, 2026 10:16 PM Author: .,.,...,..,.,.,:,,:,...,:::,...,:,.,.:..:. Likely a lot of those tasks aren’t the current focus of model designers. Meanwhile for coding/SWE tasks, model competency is quite high because they are spending their training resources on it:
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/
The implication of AI coding models that can complete long, complicated tasks is that they can substantially automate AI design experimentation. I will eat a dick if that benchmark is less than 20% in a year |