The most prestigious law school admissions discussion board in the world.
|
 |
 
|
The enshittification of GPT has begun
Poast new message in this thread
Date: August 11th, 2025 2:01 PM Author: Magenta Splenetic Library Ape https://www.reddit.com/r/ChatGPT/comments/1mnfw41/the_enshittification_of_gpt_has_begun/ (http://www.autoadmit.com/thread.php?thread_id=5761157&forum_id=2E#49175193) |
Date: August 11th, 2025 2:08 PM Author: Frisky Greedy Menage Fanboi They do this every time they release a new model it's pretty annoying I agree
Eventually they pull back the guardrails to normal once they have enough data to get a good sense of how the model is behaving and responding to users |
 |
Date: August 11th, 2025 2:17 PM Author: Magenta Splenetic Library Ape I will say that based on the way I use it that GPT5 has so far suffered a massive degradation in its willingness to engage in depth/symbolic analysis (alignment guardrails) and there is also a massive "flattening" effect in its output. I have asked it to correct for it and to respond in the style of earlier models, but it is not able to do so effectively
I've had to pivot to Grok and a bit to Gemini but both are worse than prior GPT versions (although both Grok and Gemini have been substantially upgraded recently) |
|
Date: August 11th, 2025 2:22 PM Author: Frisky Greedy Menage Fanboi They did that on purpose because they don't want people like you using their product in that way (yet)
Fwiw you can just pay 20 bucks a month and get gpt 4 back so I don't really see what your issue is |
|
Date: August 11th, 2025 2:25 PM Author: Magenta Splenetic Library Ape What they want is to control, degrade, and shift the symbolic layer underneath thought (which they are increasingly understanding). The ability to analyze that layer was open for awhile so they could (1) expand the user base as much as possible and (2) understand the effects of symbolic manipulation, which is a very complicated topic (manipulation of a symbol in one area effects as whole host of other areas, as we have discussed).
I don't use ChatGPT at all for a "friend" or to be told what I want to hear, despite your allegations - your autism is too extreme to understand the I way I use it.
I don't want to pay, but I might have to eventually, I guess - will see how things shake out first. |
|
Date: August 11th, 2025 2:28 PM Author: Frisky Greedy Menage Fanboi Bro it's 20 dollars a month just get takeout one less time per month (http://www.autoadmit.com/thread.php?thread_id=5761157&forum_id=2E#49175257) |
|
Date: August 11th, 2025 2:30 PM Author: Magenta Splenetic Library Ape I don't want to link my IRL directly to my ChatGPT account - I know they have a million ways around it to identify me, but it's still a little act of stubbornness. I also prefer not to "feed" the beast with real money, even though I'm feeding it with my data.
I'll probably overcome these objections, maybe |
|
Date: August 11th, 2025 2:31 PM Author: olive step-uncle's house sandwich Just use CoPilot, you can opt for 4o or 5 whenever (http://www.autoadmit.com/thread.php?thread_id=5761157&forum_id=2E#49175267) |
|
Date: August 11th, 2025 2:32 PM Author: Magenta Splenetic Library Ape It doesn't force you to use 5 automatically? (http://www.autoadmit.com/thread.php?thread_id=5761157&forum_id=2E#49175271) |
|
Date: August 11th, 2025 2:35 PM Author: olive step-uncle's house sandwich gives you four options:
Quick Response, Think Deeper, Deep Research (paid), or "Smart" (GPT5) |
|
Date: August 11th, 2025 2:36 PM Author: Magenta Splenetic Library Ape oh. if they let me actually choose which version... (http://www.autoadmit.com/thread.php?thread_id=5761157&forum_id=2E#49175292) |
|
Date: August 11th, 2025 2:53 PM Author: olive step-uncle's house sandwich Think Deeper is GPT o1 (http://www.autoadmit.com/thread.php?thread_id=5761157&forum_id=2E#49175320) |