The most prestigious law school admissions discussion board in the world.
|
 |
 
|
5/31/25 AI thread
| Fishy home | 05/31/25 |
| Fishy home | 05/31/25 |
| galvanic sweet tailpipe | 05/31/25 |
| Fishy home | 05/31/25 |
| galvanic sweet tailpipe | 05/31/25 |
| Pink university | 05/31/25 |
| Pink university | 05/31/25 |
| Fishy home | 05/31/25 |
| Fishy home | 05/31/25 |
Poast new message in this thread
Date: May 31st, 2025 9:59 AM Author: Fishy home https://x.com/theo/status/1928653885230497842
Ranking LLMs by how aggressively they snitch you out to the government |
Date: May 31st, 2025 10:00 AM Author: Fishy home https://x.com/ArtificialAnlys/status/1928477941715079175
I'm skeptical of this data and suspect Chinamen may be behind this "analysis"
Still food for thought |
|
Date: May 31st, 2025 10:02 AM Author: galvanic sweet tailpipe i tried using AI yesterday at work to write a Python script for me because i haven't written any Python code in years and am unfamiliar with the libraries i needed to use. it just kept giving me buggy script after buggy script, and i would feed the AI the exceptions to fix the bugs. after about 20 rounds of that i gave up. (http://www.autoadmit.com/thread.php?thread_id=5731759&forum_id=2#48976179) |
 |
Date: May 31st, 2025 10:08 AM Author: Fishy home Apparently the models are actually getting *worse* at coding now. They spit out all kinds of superfluous garbage instead of just doing what they're told
I'm not good enough at coding to know but everyone is saying that |
|
Date: May 31st, 2025 10:11 AM Author: galvanic sweet tailpipe i've been saying this ever since the first time i tried to use AI for coding. it can and does help regularly in specific, narrow scenarios, but often it just spits it buggy garbage. (http://www.autoadmit.com/thread.php?thread_id=5731759&forum_id=2#48976221) |
|
Date: May 31st, 2025 3:26 PM Author: Pink university it depends on what you want. they are getting much better at coding (compared to 1-2 years ago) if you want 1) code that will run without errors or minimal debugging 2) take a vague idea and turn it into production code.
the code they currently tend to generate is different than human generated code. it tends to include a lot of extra error handling code, or code that has redundant logic in it to prevent the possibility of errors. the code is increasingly not clean or easy to interpret. this is likely the result of more RL training. i am guessing this will only get worse before it gets better (at a certain point, the models will be robust enough that they don't need all the error handling garbage). |
|
Date: May 31st, 2025 3:27 PM Author: Pink university what model? what type of prompt were you using? (http://www.autoadmit.com/thread.php?thread_id=5731759&forum_id=2#48977172) |
|
Date: May 31st, 2025 3:16 PM Author: Fishy home https://x.com/ExaAILabs/status/1928578004458025201 (http://www.autoadmit.com/thread.php?thread_id=5731759&forum_id=2#48977109) |
|
Date: May 31st, 2025 3:20 PM Author: Fishy home https://x.com/insectbeau/status/1928709224382738654
Lmao |