The most prestigious law school admissions discussion board in the world.
|
 |
 
|
what “clicked” for ai/llms in 2022?
Poast new message in this thread
Date: August 25th, 2026 3:06 PM Author: snake coke for 30+ years, “ai” consisted of borderline useless proofs-of-concept. then, within a span of a year or two, dozens of companies were suddenly able to produce far, far more advanced and effective models. what sudden discovery/shift/dynamic allowed for this sea change? (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092219) |
Date: August 25th, 2026 3:08 PM Author: average/ordinary/typical citizen/person CERN unlocked portal to other side of black hole hell from Event Horizon for demonic possession of computer shell multiverse seepage. Why do you think it's suddenly "shut down" after all this? (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092226) |
 |
Date: August 25th, 2026 4:53 PM Author: Glengarry Glen Goy this not flame
the demons received the proper blood sacrifices imo |
|
Date: August 25th, 2026 7:04 PM Author: average/ordinary/typical citizen/person You bet your ass it's not flame bud (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092793) |
|
Date: August 25th, 2026 4:53 PM Author: Metaphysics is Goy (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092522) |
|
Date: August 25th, 2026 3:35 PM Author: The Buzz Aldrin Spacelaw Fellowship Seminal "Attention is All You Need" transformer paper in 2017. Nvidia launches V100 same year. BERT comes out in 2018. GPT comes out in 2019. Nvidia Launches A100 in 2020 (3x performance of V100) GPT-3 comes out 2020 H100 launches (5x of A100) 2022 GPT-4 comes out 2023 |
|
Date: August 25th, 2026 4:54 PM Author: Metaphysics is Goy yeah but your missing the demon and jew angles as other poasters have discussed itt (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092524) |
|
Date: August 25th, 2026 4:46 PM Author: timothy mcveigh Idk but whatever it is it happened in Israel, that much i do know (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092511) |
|
Date: August 25th, 2026 4:53 PM Author: Glengarry Glen Goy 180 monikor poast synergy (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092520) |
|
Date: August 25th, 2026 5:55 PM Author: timothy mcveigh Thanks brotha (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092655) |
|
Date: August 25th, 2026 4:53 PM Author: Metaphysics is Goy (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092521) |
|
Date: August 25th, 2026 4:57 PM Author: agenticman (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092526) |
|
Date: August 25th, 2026 7:19 PM Author: ,.,..,.,..,.,.,.,..,.,.,,..,..,.,,..,.,,. the tokenization process became faster and more workable. think of a 'classic' AI program like Dr. Sbaitso. it didn't really have a 'transformer' process behind its output. it mostly just took inputs and spat out a reply based on a lookup table of pre-programmed responses.
a modern Dr. Sbaitso would use a different system based on abstract representations of user inputs. it wouldn't just use a basic parsing tool to look at the text; the input text would be converted into a series of numerical values/tokens. these tokens would be placed into a mathematical vector space in relation to other tokens, which is why graphics cards are so important to this kind of processing.
modern Dr. Sbaitso would run the token through the vector space and then reconstruct a reply. it's no longer a direct Input(A) --> Program(B) --> Output(C) model. everything flows through the vector space before an output is rendered. |
|
Date: August 25th, 2026 7:21 PM Author: Founder scale (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092836) |
|
Date: August 25th, 2026 7:21 PM Author: Glengarry Glen Goy better demons and better sigils (http://www.autoadmit.com/thread.php?thread_id=5897084&forum_id=2/#50092839) |