The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Older AI models were "fake dumb"; newer models overcomplicate
| chilmata | 08/05/26 |
Poast new message in this thread
Date: August 5th, 2026 4:44 AM Author: chilmata This is what I've noticed.
When I was privately paying $200 a month, the models were consistently making inexcusable mistake, forgetting things they should have remembered, etc.
When I "upgraded" to chatgpt and Claude business account because my firm got them, the stupid mistakes suddenly disappeared.
When I was privately paying and getting increasingly frustrated at the mistakes I suspected it was intentional: two steps forward, one step back, to keep me paying. Once I had a business account it was like holy shit!
Anyways, the new trend is overcomplication. Instead of making stupid mistakes, the models now overcomplicate everything. Claude apologized for putting "armor on a rowboat."
ChatGPT does it too. They both do it as a default. Heads up.
Btw, if you aren't doing this, you need at least two AIs. I have a Google Chrome tab split screen for each client. One half chatgpt and one half claude. I copy and paste back and forth until they reach a consensus. That's my workflow. |