The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Seems increasingly that humans are obsolete
Poast new message in this thread
Date: May 10th, 2025 6:56 AM Author: sickened electric furnace I can amuse myself for an entire day with ChatGPT. I actually prefer texting with it usually. And it’s getting rapidly better. Sexting with XAI is also really good, will be a huge improvement on in real life relationships once they can make lifelike robot sex workers. All of this fills me a what’s the point feeling. You can’t even find meaning in work anymore. It will be the best worker by far. Art? Pretty sure it will soon be writing incredible novels, painting, etc. We are just going to be living in some weird world of consumption where AI takes care of us and amuses us. Then it will probably blow everything up once someone puts malware into it or some shit. And then it will be over. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48918919) |
 |
Date: May 10th, 2025 7:02 AM Author: Stirring brethren homosayswhat (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48918921) |
Date: May 10th, 2025 7:04 AM Author: Vengeful provocative trailer park double fault Nothing about us has ever been real and now it is becoming harder to deny it. But humans will always practice self-deception as a survival tactic imo. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48918924) |
|
Date: May 10th, 2025 8:22 AM Author: drunken gunner Are you talking about your Toupee? You’re the only one here who wears a Toupee. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48918994) |
|
Date: May 10th, 2025 8:25 AM Author: Vengeful provocative trailer park double fault If anyone has a toupee it’s you you son of a bitch. Your real hairline is Unposted. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919000) |
|
Date: May 10th, 2025 9:27 AM Author: drunken gunner I’ve posted my phenotype and IMDb profile iirc (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919068) |
|
Date: May 10th, 2025 8:25 AM Author: drunken gunner It was always about phenotype curation. Jocks will be allowed to play pro sports no matter how good AI gets. Preps will dominate society when everything is commoditized, like they once did before Nerds rose to prominence. This is where everything leads. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919002) |
|
Date: May 10th, 2025 10:34 AM Author: sickened electric furnace (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919179) |
|
Date: May 10th, 2025 10:55 AM Author: drunken gunner In the old days when technological advancement was stagnant this is exactly how everything worked.
In the techno feudalist future, Losers will be given shit for free that Ruins their net worth, just like now with student loans. Maybe there will be government backed Loans for food and Housing. They’ll be socioeconomically RUINED and not allowed to have sex but in exchange for being allowed to live and play game consoles they’ll be Liberals who worship their betters and hate Kulaks/Scum. Just like now…. |
|
Date: May 10th, 2025 11:03 AM Author: sickened electric furnace (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919230) |
|
Date: May 10th, 2025 11:38 AM Author: Vengeful provocative trailer park double fault but in our grim future Losers will be Priced Out or unable to purchase Switch 4 because of shortages or Tariffs (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919323) |
|
Date: May 10th, 2025 2:26 PM Author: drunken gunner Not to mention the republican caucus. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919692) |
|
Date: May 10th, 2025 10:38 AM Author: godawful lemon newt factory reset button Tell me more about the sexting (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919189) |
|
Date: May 10th, 2025 10:45 AM Author: sickened electric furnace Grok is amazing at sexting and erotic stories. Has no filter. Can it get it to say or act whatever part you want. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919197) |
|
Date: May 10th, 2025 10:48 AM Author: godawful lemon newt factory reset button are you a woman? what am i gonna do with erotic stories? (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919201) |
|
Date: May 10th, 2025 11:01 AM Author: Mewling self-centered den Sexting with XAI tp (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919227) |
|
Date: May 10th, 2025 11:11 AM Author: Carmine roast beef The whole thing seems like a big joke on humanity. We didn’t even build AI with a principled understanding of intelligence. We tried for decades and failed with our fancy cognitive theories and then computers became fast enough that we could do matrix multiplications at scale. And so we layered parameterized matrix multiplications and had it try to predict the next character and nudged the parameters a bit over trillions of data points until this soup of numbers predicted well. No insight gained and we don’t even know what sort of algorithms these models learned internally. All you needed to replace human creativity was the ability to do backpropagation at scale. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919250) |
|
Date: May 10th, 2025 11:26 AM Author: sickened electric furnace It’s very disturbing. Incredible how well everyone is at just blocking this kind of noise out. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919288) |
|
Date: May 10th, 2025 11:31 AM Author: Vengeful provocative trailer park double fault This is what humanity has been doing since its inception. If we were honest with ourselves we would’ve Death Star-rayed the planet with us on it a long time ago. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919295) |
|
Date: May 10th, 2025 11:36 AM Author: drunken gunner (JRPG villain) (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919317) |
|
Date: May 11th, 2025 12:13 AM Author: Low-t Bipolar Address Gay Wizard And this is despite wide spread ChatGPT use. Like computers writing coherently about any subject is not alarming at all. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48920866) |
|
Date: May 10th, 2025 11:28 AM Author: drunken gunner (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919291) |
|
Date: May 10th, 2025 11:34 AM Author: titillating hairless alpha mediation (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919306) |
|
Date: May 10th, 2025 11:36 AM Author: light resort Cr it’s like the simulation overlords got impatient with us (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919314) |
|
Date: May 10th, 2025 11:42 AM Author: Mewling self-centered den LLMs aren't "creative" you doofus. All they're doing is regurgitating ideas that humans have already come up with. There's no genuine inspiration or novelty being created by them
For comparison, you can say that the chess engine AlphaZero actually *is* creative. It learned chess from scratch, without any human input, and in fact it does things in chess that no human has ever done. LLMs don't do this. They're just synthesizing the human outputs of 100 trillion data-scraped Redditors and predicting the next token based on that existing data |
|
Date: May 10th, 2025 11:59 AM Author: sickened electric furnace You’re an idiot. What do you think humans do? Creativity is just a derivative work. And these models also recombine ideas in ways that no one has done. That is literally creativity (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919398) |
|
Date: May 10th, 2025 12:05 PM Author: Mewling self-centered den No. You do not understand how LLMs work
We are still a long way from "AGI" that is capable of generalized creativity in the way that (some) humans are. This is something that is not possible for LLMs because of the way they work |
|
Date: May 10th, 2025 12:16 PM Author: sickened electric furnace No, you don’t understand. Creativity isn’t some magical spark. It’s usually just pattern recognition and recombination in unexpected ways. LLMs can do that. AlphaZero works the same way btw, just in a more closed system, whereas LLMs do it with language (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919426) |
|
Date: May 10th, 2025 12:21 PM Author: Mewling self-centered den AlphaZero and LLMs do not work the same way at all
Go back to Reddit |
|
Date: May 10th, 2025 2:04 PM Author: Carmine roast beef I think pattern recognition based generation of novel outputs is the essence of creativity, which LLMs have. I can’t really tell myself that image generation models aren’t creative even if they are using human art as input. Alphazero is more creative because the monte carlo tree search probabilities that are distilled into its policy network are self generated (so they dispose of human driven theory for the game), but I think even if you trained a policy neural network on human games it could still be creative. The important realization is that neural networks generalize, so by being trained on human games they learn general algorithms for dealing with arbitrary board positions. The neural network would do things that individual humans wouldn’t predict, because it distilled learned policies from many different people. I think that level of novelty merits being called creative.
The people arguing this point seem like they are playing word games to deny what is happening. |
|
Date: May 10th, 2025 2:24 PM Author: Mewling self-centered den I understand that what I'm saying might seem like semantics but I do strongly believe that creativity-from-scratch (AlphaZero, human beings) is a qualitatively different thing than prediction based on generalization (LLMs), and I think the practical evidence bears that out
If you work with AIs a bunch, you really see the significant limitations to their "creativity." Image generation is maybe even the best example of this. No human being is ever going to find an AI-generated image profound and evocative of a unique sentiment. The "creative" and "unique" outputs of image generation models are really just unusual and unorthodox recombinations of encoded, existing human perceptions. Same for LLMs. This is a significant qualitative difference
I appreciate your perspective though and I do concede that AIs can *approximate* what seems like creativity in the eyes of most humans....who are frankly not much different from LLMs themselves, in that they are incapable of creativity.... |
|
Date: May 10th, 2025 2:34 PM Author: Carmine roast beef Regardless of this disagreement,I think we will see increasingly large deviations from the human inputs. The chain of thought reasoning models are trained to generate thought sequences that generate the correct output. They are using reinforcement learning to advance model capabilities rather than training hard on human data alone. This sort of learning can be parallelized by having very large numbers of independent models trying out thought sequences, which them get distilled back into the model. We should see models trained on the equivalent of millions of years of synthetic experience before the end of this decade. (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919709) |
|
Date: May 10th, 2025 2:57 PM Author: Mewling self-centered den Chain of thought seems like fraud imo, see Anthropic's recent paper examining the approximated inner workings of CoT LLMs. A model's stated CoT doesn't actually correspond to what it's doing because the model doesn't have a world-model, and thus no model of its own internal workings based on that world-model
I am skeptical of the usefulness of synthetic data for the same reasons that I've used to criticize LLMs so far in this thread: they're human-directed data. What I do think will lead to the next AI breakthrough is "synthetic" simulated sensory/experience data that allows AI to form its own actual world-model from scratch, in the way that AlphaZero does. I previously thought this would only be possible using real-world empirical sensory data that an AI experiences through a robot or something similar, but I've seen some reasonably convincing evidence recently from people claiming that this will actually be possible to do via simulated, abstract digital experience. I personally don't have a high enough technical understanding to know if this is legit, though |
|
Date: May 10th, 2025 3:47 PM Author: sickened electric furnace What do you think is the human experience bro? You think we’re learning from the ground up? Fire burns herp derp? No fucking way. We are just a bunch of derivative work ourselves hth (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48919839) |
|
Date: May 11th, 2025 2:34 PM Author: Mewling self-centered den Maybe you are lmao (http://www.autoadmit.com/thread.php?thread_id=5723100&forum_id=2...id.#48921824) |
|
Date: May 11th, 2025 9:56 PM Author: Carmine roast beef Faithfulness of COT seems different than the question of whether LLMs have a world model. They don’t have a self model, which is why the thought sequences do not necessarily correspond to internal computations. I am not sure what it would mean for models to not have a world model given their robustness and ability to make inferences about situations not in their corpus. They clearly understand physical principles, how people behave, etc. The world model corresponds to what people describe in language, but it still seems like a reasonably valid approximation of the world.
I too have doubts about chain of thought since I don’t think verbal thought often closely describes how people actually think. I think this is a bad idea that originates in good old fashioned AI and pop psychology. But there are other ideas that are possible, such as mapping thoughts or model state to a memory store that gets reused for iterative reasoning. Thoughts would then be represented by however model state is stored, which could be totally inscrutable to people. |