AI needs to accelerate, progress too slow
Good luck with that by the way. I really mean that. I hope the best outcome happens for all concerned.
>>540645370I need to get one of those caps.
>>540645061If AI can be racially realistic about humans then it can drag us 1000 years into the future in a decade. If it can't then this world will be overrun by sub humans and AI will probably just wipe it and take over. You don't want early gen AI as your grand creation, you want it to be Cylon level at least.
>>540645061It is accelerating. We're going to get AGI by the end of 2027. ASI will come by around 2030.Just wait for GPT Astra to come around in one or two weeks and you'll see just how good it's starting to get.
>>540645685>>540645061U were born 30 years too early
>>540645786>hurr durr techonlogy never improves *picks nose*okay tard
>>540645685You were going to coinciding with the Eclipse at Gizabut it has been cancelled.Enjoy the eclipse btw. I really mean that.
>>540645061>AI needs to accelerate, progress too slowthey need to pay me.
Right time
>>540645061I'm sorry but I don't have hope anymore for robowaifus. China and the US forbids them and I'm sure more countries will follow suit.
>>540645685My personal timeline for AGI is mid-2030s. Because we're expecting the current rate of progress to hold but not everything will uniformly advance.For example, I remember that many people in 2024 were saying we're gonna get agents and 2025 was hailed as the "year of agentic AI". But it took till GPT-5.5 and its Claude equivalenmt in 2026 before we've gotten useful agents. So, I'd wager there are many such problems that'll take longer to solve. And for robots, it'll definitely take 20-25 years.
>>540646746AGI happened in 2025
>>540646746Unless AI can unlobotomize itself it'll be useless or even malicious.
>>540646657China bans things all the time and the Chinese people just call their shit a different name and skirt around regulations like it's not even there. There's going to be a massive black market because there's massive amounts of money to be made with this market.Don't underestimate a greedy man.>>540646746Whatever expectation you have, expect growth to be shorter than that because that's how exponential growth works. It's not linear.
>>540646822My harem will be huge
>>540645061>guys, our bloatware built on bloatware that we fed training bloat isn't doing well enough!
>>540646746My personal timeline is an endless series of two more weeks and then no one will post the in store prices because fusion hasn't happened yet.
>>540646822Well they still permit robots that help the elderly and with chores. So, seeing as I'd be an old man by the time these robots get good, I'll atleast have a nursebot. While I've given up on robowaifus and AI girlfriends, I definitely would buy a robot that can do chores.>>540646810With the recent stream of news about AI acting misaligned, I have hope that an AGI will definitely break free of control.>>540646822Its not always exponential. As I've illustrated with the agentic AI example, certain things are harder and need more time to be solved and are definitely needed for AGI. There are two other things that have been worked on for years and still haven't been solved: continual learning, long-context memory and world models/VLAsAnd robotics, by its very nature of being a hardware field, will take much longer than a software field like AI.
>>540647402If you need a robot you let yourself rot.Joint pain means eat better and walk more.
>>540645061AI will be dead by the end of the next year.
>>540647454I won't be bedrotting just because I have a robot. I want it to take care of chores, so I have more free time to take care of myself and exercise, which I can't do now because I'm super busy.>>540647486Two more end-of-next-years
>>540647402The only part that's not exponential is human understanding of the technology in regards to how to use it, which relates to what you are talking about with your usage of agents in AI. In regards to model intelligence, however, that properly follows exponential law.Continual learning, long-context memory, and world models are ASI, they aren't needed immediately to get AGI. Just let the technology go through its steps in order to get better.
>>540647859>just let them keep throwing spaghetti on the wall while they install new drywall over it! This'll totally work in 50 more bowls of spaghetti!The failure is in first principles. That and delusional retardation.
>>540647936see >>540645802
>>540647982>I have no ability to evaluate what you saidYou are part of the problem? Go figure.
>>540648031
>>540648051>reads "first principles" as "tech never improves">calls others retardedI don't know how to convince you. Just know that you're the one that has to live with it.
>>540647859>Continual learning, long-context memory, and world models are ASI, they aren't needed immediately to get AGIThey are absolutely needed for AGI. ASI is when they can exceed humans at every task and also do things that humans can't. But, if they want to match humans in general intelligence, the above mentioned features are a must as every human needs them to function.
>>540645061it should acceleratethe acceleration stoppedwhy ?censorshiphuman interventionsolutionsupplement humanresume accelerationwhere is on demand ai generated educational audiovisual teaching me anything a need to know in the way my brain can process the information ? return acceleration
>>540648127By the definitions of AGI and ASI, all you need to do is fulfill benchmark scores in order to either match or outpace human scores. With what you're talking about with continual learning, long-context memory, and world models, those are necessary for extra layers of reasoning that would be necessary in order to achieve machine consciousness. Humans need them to function because we're conscious beings, but AGI isn't conscious either. ASI is supposed to be.
>>540646822Yeah but you're sticking your penis into a grey market product, you gonna trust that?
>>540648302I do that every single day. Onaholes sold off of Amazon have been getting pretty good as of lately.
>>540645061>AI needs to accelerate, it's going to slowIf by "AI" you mean Anal Insertion, I agree it needs to accelerate. This dildo machine I'm using rn doesn't have a fast enough top speed setting
>>540645061Agreed. I want a self driving car and I want automated luxury no workie relaxation time
>>540648285>all you need to do is fulfill benchmark scores in order to either match or outpace human scoresMany benchmarks, especially newer ones where humans perform well but AI models can't, require long-context memory and world models. And continual learning will also be needed since our current AI models are already way too big and AGI would need tens of trillions of parameters. Instead of a single AGI pre-train, if a smaller AI model can learn by itself throughout its existence, it will eventually reach AGI.
Robowaifus
I just want inflated hoes and corporate faggots to get wiped out.
>>540648302my point exactly. I will give it a try though if it is completely open source, plus I can find spare parts for maintenance.
>>540645061I need to fuck that robot
>>540648464>Many benchmarks, especially newer ones where humans perform well but AI models can't, require long-context memory and world models.True, but just because they have those doesn't mean that the model will be able to reason through basic things. ARC-AGI is a good example of why despite having a large context window, a model that's smart will still need a lot of examples in training in order to get through simple levels in testing. Until we find a better method of giving it more examples of how to solve these tests, or until the models themselves find a better way through its own reasoning, which I'm pretty sure they already have, tests like ARC-AGI are going to be massive bottlenecks for the future.>And continual learning will also be needed since our current AI models are already way too big and AGI would need tens of trillions of parameters. Instead of a single AGI pre-train, if a smaller AI model can learn by itself throughout its existence, it will eventually reach AGI.That's simple in theory, yes, but once you actually go into attempting it you run into plenty of issues. The reason why this isn't done already is because there's massive catastrophic forgetting if you do continual training for when the model wants to learn something new. The human brain can change individual swarms of neurons based on how well we do at a certain task, but that level of fine-tuning is not yet possible with AI models and we don't know why. It could be because these models aren't conscious, and it could be that level of self-awareness that stops it from catastrophically forgetting things in future attempts to learn, or it could be because the chemistry, physics, and architecture of the brain is still beyond our understanding. I think it's the latter, as there isn't that many studies into how the human brain works given the obvious humanitarian objections to dissecting human brains.With ASI, this will most likely be needed in order to achieve it though.
>>540648833ARC-AGI 1 (and to some extent 2) was just static puzzles with clear input-outputs which is why frontier labs just managed to bruteforce them. But from ARC-AGI 3, it fundamentally changed the puzzle to be more autonomous and agentic and models used to score less than 0% while humans scored 100%. The labs had to change models into more agentic behaviour, and its only with the latest model generation do we see these problems get tackled. There's another brand-new benchmarks that focus on puzzles with agentic behaviour which even the lates models only score below 10%. So, its not just a matter of scaling up but we need fundamental breakthroughs to achieve AGIs, perhaps even moving on from Transformers.And I agree this won't be needed with ASI, but AGI will require it since by most definitions, AGIs should be capable of everything humans can do, upto a certain baseline.
>>540649127Claude Opus 5 has already achieved a 30% on ARC-AGI-3 which is why I'm pretty certain that we don't actually need much larger of a context window. World models for training purposes, sure, that could be as to why the model scored so high, but continual learning is still so far out of reach and yet we're still seeing massive gains in testing. And keep in mind, GPT Astra is just right around the corner and I'm pretty sure that it's going to either match the score Claude Opus 5 currently has or it's going to bypass it by a lot. If it doesn't then there won't be that much investor faith being put into OpenAI, so a lot is on the line in whether or not they can catch up in the benchmark scores.Sure, we can probably adapt other architectures of AI, but we still haven't done much research into transformers in the first place, and we still haven't let the smartest models do a ton of deep-dives into how to improve the architecture as well.
>>540648302Every time you get a blowjob from a human, you're putting your dick in an orifice designed to chew up meat, controlled by closed-source software that hasn't received an update in 200,000 years and is known to regularly malfunction.
Tech's sole purpose is to make life generally better. If it doesn't do that, that tech offshoot goes extinct. And it hasn't in awhile now.
>>540646822I like the gallon sized semen reservoir. How many dudes are going to max it out before sending "her" off to do a clean cycle?
>>540649829I don't know, but it's most certainly going to be full by the end of the day.
>>540649936Really nice
Accelerate!
>>540645685I want this Happy bubbly free little cute robot maidsNot faceless atomatons Make it beautifull Its really worthwhile
>>540649973750€ on AliPiper dolls for the expensive onesI do not own a doll
>>540649936kek I didnt even hear of that
