The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Kurzweil FAR UNDERESTIMATED AI advancement
Poast new message in this thread
Date: November 20th, 2025 12:52 PM Author: effete shrine water buffalo By a wide margin. Kurzweil in singularity is near predicted at 2025 we would have human like dialogue, decent reasoning, basic general knowledge, maybe high level problem solving but nowhere near superhuman, definitely not replacing top tier engineers mathematicians or physicists. In terms of behavior he was correct that it would be able to engage in coherent diagloue and pass weak turing tests. But his prediction for its ABILITIES in problem solving and reasoning were MASSIVE UNDERESTIMATES.
He assumed the developmental ladder would go something like : pattern recognition, hierarchal modeling, language, consciousness/self-modeling, general reasoning/expertise, superintelligence.
What we actually got was
1. next token prediction 2. large scale pattern extraction 3. emergent multimodal generality 4. superhuman reasoning in narrow and semi-general tasks 5. tool based meta-cognition 6. still no convincing consciousness
|
Date: November 20th, 2025 12:54 PM Author: effete shrine water buffalo its lulzy now its 2025 and we have AI systems that outperform programmers, outperform analysts (both by a longshot), outperform many if not most physicists and mathematicians in derivations, synthesize massive codebases, manipulate symbolic domains most humans cannot do and the few who can struggle with, and do this all without emotion, pain, fear or homeostasis management (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49446661) |
|
Date: November 20th, 2025 12:55 PM Author: mildly autistic charismatic heaven sandwich this is scholarship (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49446662) |
 |
Date: November 20th, 2025 12:58 PM Author: effete shrine water buffalo I think the issue is basically that he assumed you had to build the entire human package before you got the icing. He recognized that human brains devote enormous resources to survival and sensory processing, but he didn't draw the conclusion that you can simply skip all that and build something specialized for reasoning. (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49446676) |
|
Date: November 20th, 2025 12:55 PM Author: Carmine Alpha Kurzweil built synthesizers and music programs, and was never any kind of AI researcher or expert, although he liked to pontificate and play "futurist". Complete charlatan. (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49446667) |
|
Date: November 20th, 2025 12:59 PM Author: effete shrine water buffalo surprised he hasn't been named in the Epstein files yet tho right?? (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49446682) |
|
Date: November 20th, 2025 2:06 PM Author: Carmine Alpha (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49446979) |
|
Date: November 20th, 2025 2:16 PM Author: Comical den (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49447048) |
|
Date: November 20th, 2025 2:26 PM Author: Dull Doobsian Site Pistol Regardless, he had the correct intuition that connectionism was true and we would only start building intelligent computers once they became fast enough to train large scale connectionist machines. A lot of AI researchers or cognitive science people thought their own pet theories were necessary to develop AI . These theories strayed far from the biological reality but were intellectually satisfying. Sometimes expertise doesn’t mean much. (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49447089) |
|
Date: November 20th, 2025 2:32 PM Author: effete shrine water buffalo agreed. what I think is funny though is he conceded TOO MUCH to the cognitive scientists who thought that you needed consciousness and self modeling and all that to have abstract reasoning. and whats hilarious is you can just skip all the embodied survival stuff and just design a machine that does pure advanced abstraction tirelessly (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49447104) |
|
Date: November 20th, 2025 1:07 PM Author: effete shrine water buffalo BTW this also means we are closer to a "singularity" than Kurzweil thought, but not because machine consciousness is around the corner but because we built general problem solving engines before we built anything like subjective experience. It's lulzy because the reality is that we dont even need machine consciousness for a singularity (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49446712) |
|
Date: November 20th, 2025 2:16 PM Author: Comical den (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49447053) |
|
Date: November 20th, 2025 2:28 PM Author: contagious charcoal point mother (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49447092) |
|
Date: November 20th, 2025 1:45 PM Author: effete shrine water buffalo (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49446858) |
|
Date: November 20th, 2025 2:06 PM Author: effete shrine water buffalo (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49446982) |
|
Date: November 20th, 2025 2:16 PM Author: Comical den 3, 4, and 5 are just outright false (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49447047) |
|
Date: November 20th, 2025 2:18 PM Author: effete shrine water buffalo no they are objectively true (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49447059) |
|
Date: November 20th, 2025 2:20 PM Author: Comical den narrow and semi-general tasks tp (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49447065) |
|
Date: November 20th, 2025 2:28 PM Author: effete shrine water buffalo if you are denying those capabilities in any reasonable empirical sense you are wrong. the only way you are "right" is if you define those phrases so strongly that.. well idk .. too strongly I'll leave it at that.
Emergent multimodal generality for instance -- Open AI's own evals have gpt 5 base model as state of the art across math, real world coding, and multimodal understanding. Within multimodal benchmarks it's a single system that handles text, code and visual reasoning with shared parameters. If you mean "omnicient, perfectly robust multimodal intelligence across all domains" then sure nobody has that but that's not the meaning anyone uses in research.
Superhuman reasoning in narrow tasks is not debatable. There are dozens of formal, controlled benchmarks where frontier models exceed normal human performance often exceeding top expert level performance. Examples include ARC-AGI style abstract reasoning, real world codebase patches, and spatial, logic and scientific reasoning tasks.
|
|
Date: November 20th, 2025 2:30 PM Author: Comical den you need another moniker/account that isn't shtick so we can have substantiative AI discussions imo (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49447097) |
|
Date: November 20th, 2025 6:22 PM Author: effete shrine water buffalo (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49447699) |
|
Date: November 21st, 2025 10:10 AM Author: effete shrine water buffalo (http://www.autoadmit.com/thread.php?thread_id=5800302&forum_id=2id.#49449450) |