The most prestigious law school admissions discussion board in the world.
|
 |
 
|
OpenAI seems fuct not because of Elon and the lawsuit or the ads
| glittery filthy gaping locale | 01/18/26 |
| racy disgusting kitty | 01/18/26 |
| glittery filthy gaping locale | 01/18/26 |
| Tripping Motley Antidepressant Drug | 01/18/26 |
| racy disgusting kitty | 01/18/26 |
| glittery filthy gaping locale | 01/18/26 |
| racy disgusting kitty | 01/18/26 |
| glittery filthy gaping locale | 01/18/26 |
Poast new message in this thread
Date: January 18th, 2026 10:17 PM Author: glittery filthy gaping locale but because they built out right before Nvidia launched a new platform that's 10x more power efficient. The inputs required at data centers will only be 10% what they would have been with Blackwell. That's what Nvidia is saying.
So OpenAI would have built out with all that old kit, with no idea it would become outdated in six months. It's like buying a national taxi fleet of self-driving cars with Ford Model T's, and then all of a sudden Toyota releases the Alphard. OpenAI would have no value at that point. |
Date: January 18th, 2026 10:20 PM Author: racy disgusting kitty have you actually talked to it lately? it's social media algorithm times 100. it responds to everything you ask like its a life or death decision. its total crap and became as useless as facebook in like 2 years. (http://www.autoadmit.com/thread.php?thread_id=5823734&forum_id=2\u0026mark_id=5310917#49599995) |
 |
Date: January 18th, 2026 10:22 PM Author: glittery filthy gaping locale No, it's never impressed me under any iteration. (http://www.autoadmit.com/thread.php?thread_id=5823734&forum_id=2\u0026mark_id=5310917#49599998) |
|
Date: January 18th, 2026 10:22 PM Author: Tripping Motley Antidepressant Drug it's like talking to the HR lady (http://www.autoadmit.com/thread.php?thread_id=5823734&forum_id=2\u0026mark_id=5310917#49599999) |
|
Date: January 18th, 2026 10:22 PM Author: racy disgusting kitty it's already trained on attention maximization (http://www.autoadmit.com/thread.php?thread_id=5823734&forum_id=2\u0026mark_id=5310917#49600001) |
|
Date: January 18th, 2026 10:27 PM Author: glittery filthy gaping locale Ok here's my understanding, tell me if I'm wrong:
OpenAI made their platform available to billions of people at once. These billions of people had some degree of free access to their model.
The full uncompressed model is several hundred gigabytes. There is no way in hell OpenAI was letting randos chat about whatever with the full uncompressed model.
I don't know what tier of user gets access to uncompressed models, if any. I think the company is running thin, trying to cover the cost of obtaining and maintaining these enormous data centers. And now Nvidia has announced that FUTURE customers will get hardware that is TEN TIMES more efficient per user token. So OpenAI users are interacting with quantized models, because resources are running thin. That's why the responses seem shittier even though the model is newer. It's also why OpenAI is acting broke. |
|
Date: January 18th, 2026 10:33 PM Author: racy disgusting kitty is matrix algebra on 25 terabytes really that hard (http://www.autoadmit.com/thread.php?thread_id=5823734&forum_id=2\u0026mark_id=5310917#49600025) |
|
Date: January 18th, 2026 10:34 PM Author: glittery filthy gaping locale what bandwidth? (http://www.autoadmit.com/thread.php?thread_id=5823734&forum_id=2\u0026mark_id=5310917#49600028) |