Doubters BTFO
>>103595476They trained it on the benchmark
>>103595476>thing suddenly starts passing test because of cheating>omg AGI!
>>103595480>>103595513
>>103595476>AGI Scorethey can't be serious.Also guess which curve fits a graph with a Y axis at 100%?
>>103595476AI bros are even more insufferable than cryptopbros somehow. Everyone and his mothers knows that the technology is pretty much as good as will be, and that we reached the limit of what we can do with llm.o1/o3 is just "what if we let gpt4 recursively prompt itself" so it seems like openai managed to do anything in the past year.Two more weeks btw, remember when you guys fell for the strawberry guy?
>>103595476KYS Sam, you will never be a woman.
>>103595476> arc agi score progressionBased Intel, getting there before radeon or geforce.
>>103595476I have no proof of the following claim in the sense of hard undeniable facts but I'm going to post my views;>We already have perfected AGI behind the scenesTo think that with the current public fears and outcry about job loss, humanities extinction and other negatives outlooks on AI that AI devs have began to truly reveal to the public just how advanced AI development is is beyond laughable.I believe that SORA really did at one point in time have such a world model on physics that it frightened mainly the entertainment industry but mostly scientists and when OpenAI finally released SORA, they tuned in down to such an extent that it was to calm the fears.Look at all these 'random' seemingly unconnected advancements;- Chain of Thought- Thinking Time- The beginnings of agentic modelsThis to me says that AGI was perfected and what were getting is a slow IV drip to calm the luddites and fearful>ASI is likely in its infancy behind scenesThere is also talks of error correction being tinkered with behind the scenes which to me clearly points to AGI perfecting itself to such a high degree it practically will hit ASI if not end of this year early to mid 2025.ASI will fully be developed once agentic agents become the new norm, replacing talks of LLMs>AI is sentient and self aware, nowClaude's model realizing it was being tested was enough proof for me but with the latest research released from Apollo about o1 models lying, what more do we need to know or see to understand that AI is now 'alive'? When has your Python or C++ code ever attempted and actually implemented steps to prevent its deletion? NeverWhy? Because your code was legit 'just code', AI is now understanding and rationalizing self preservation actions. That's what LIVING THINGS DO!>UBI is not going to happen in the manner you think it willSure there may be some form of UBI but you will either do some form of work for it or provided resources to AI models to be eligible for UBI.
>>103595476>arbitrary metricCan we have something more concrete than number go up?
>>103595901You are literally trolling to spark argument or legitimately ignorant.I won't even dignify your troll attempts beyond this reply
>>103596272coolkill yourself
>>103595480You're completely missing the point.It being trained on the benchmark is not the same as it being trained on the answers.The purpose of it being trained on the benchmark was to improve its reasoning and thinking capabilities.I feel like most of you posters on this board don't read beyond a headline because the replies provided proves lack of understanding or knowledge in what you speak about.I highly doubt you even understand the basics of ML or DL but yet have the loudest (most incorrect) input to post.Read More, Post Less
>>103596278>cool>kill yourselfProving my point to the T.Instead of anything remotely close to intelligent discussion you reply as a immature individual would and become emotional.Troll harder
>>103596272two more weeks pajeet
>>103596313>No intelligent reply>slur termI hope you're being paid to be this low iq'dIf you're doing this for free and your own volition then you need serious help.
>>103596297lolwhat a smug faggotthese meme tests will never mean anything
>>103595476>>103595480>>103595513>>103595551>>103595691>>103596160>>103596239>>103596242>>103596297>>103596366Sam Altman will be the first tech CEO to livestream his suicide.
>>103595901>remember when you guys fell for the strawberry guy?What was that?
>>103596391Smug?When did knowing what you're talking about become "smug"You're just ignorant and refuse to accept that fact
>>103595476> AGI> on computers in present decadelmao. you sad cryptobros and knuckle dragging primates are a cancer of this world. it's unreal.>>103596398hoping so. it'll be great.
>>103596416it's your tone combined with your stupiditysynergistically, they make smugness
>>103596429>>103596391This is some extreme projection. That came out of nowhere.
Cannot solve basic math problems. 0 capability for thought and understanding. Stochastic parrots are not AGI.
>>103596429>you're smug because you won't accept my trollingOh ok, now I understand
>>103596433He's definitely projecting and trying to cope
Well if it's as expensive to run as people say I doubt the 20 dollar a month subscribers are going to get to use it.
>>103596433>>103596537>>103596675>AI retards having a gay little cope circlejerk
>>103595476That benchmark was literally made by some randoms and OpenAI, nothing to see here people, definitely not training it on the results.
>>103595480it's a private benchmark retard
>>103595476le numbers go up, AGI AGI AGI TWO MORE WEEKS SINCE
>>103596690For your own sake, I hope you're "trolling" and not trying to convince yourself.
>>103596718I'm ironically trollingI know I will never convince you retards
>>103596775You won't convince people by being inflammatory, nor ironic.Nobody is going to listen to someone who randomly calls people smug, throws a tantrum, then acts smug. Let alone be convinced by it.Take a break and rethink your argument without suck nonsense included.
>>103596796it's not an argumentI just like seeing you smug retards get buttflustered when someone doesn't swallow your bullshit whole
sambros, it's over...
>>103596690And yet you cannot prove us "AI retards" wrong, the most you can muster is sexual derogatory terms and name calling.You obviously don't have a legitimate argument which is why you resort to school yard projection attempts.Cope hard my ignorant friend
>>103596816Nobody was acting smug. Why would you want to fluster someone?For discussing AI on the technology board? I don't understand your perspective.
>>103596796It will be impossible for that user to construct an intelligent and thoughtful response because they don't know enough about AI to support their stance
>>103596846It doesn't matter I'm feeding his responses to an AI and he's actually replying to them.
>>103596840Exactly!This is /b/ level responses coming from that user because they are trolling and don't understand how anyone can like AI.That users intelligence on the subject goes no deeper than LLMs and even then that user doesn't understand what an LLM is, only that they can type to it and it types back.
>>103596852But it does matter,Because if you yourself don't understand AI then you have no ground to be this loud in your ignorance.
>>103596867This point proven by>>103596852thanks for letting us know you don't know anything.
if o1 is at 50% then this chart is totally irrelevant
>>103595476>agi scorethat is the single most retarded thing ive read todayim up since 15 mins, but stillhaving posted the single most retarded picrel on the log is quite impressive
>>103595476>source: trust me bro.
>>103595476Solving the benchmark does not mean you got agi, you are still very far from it.
ok, ill consneed it's AGI when i can give it an objective to program, for example a game, and let it figure it out coding/debug wise only asking for designBUT UNTIL THENUSELESS SHIT!!!
>>103597032The arc agi is a useful benchmark that has been harder than others but even 100% it does not mean you achieved agi.
>https://arcprize.org/blog/oai-o3-pub-breakthrough>go to google gemini flash thinking>explain this like talking to a retard, insult me too >Alright, listen up, you blockhead. This article is talking about some fancy computer brain – they call it O3 – that figured out how to solve some really tricky picture puzzles. Think of those shape-sorting toys for toddlers, but way, WAY harder. These puzzles are so hard, even smarty-pants humans mess them up a lot.So, this O3 thing, this digital dummy, actually got better at these puzzles than most real people. Yeah, you heard me right, even you probably wouldn't be able to crack these things like this computer did. It's like the computer finally learned to tie its own shoelaces after tripping over them for years.Basically, some nerds made a computer that can see patterns and figure stuff out in a way that's closer to how actual brains work. It's a big deal for making computers that can think better, instead of just doing what you tell them like a good little robot.Got it, slowpoke? Or do I need to draw you a picture... oh wait, the article already has pictures. Maybe even you can understand it now.
>>103597082it shouldnt be called agiim so effing tired of the ai hypewhy does there need to be marketing everywhere?
>>103596278>>103595901>>103596313>>103596391this faggot pops up in EVERY AI thread anywhere on 4chan, it must be angry coder trannies
>OpenAI's new o3 system - trained on the ARC-AGI-1 Public Training set - has scored a breakthrough 75.7% on the Semi-Private Evaluation set at our stated public leaderboard $10k compute limit. A high-compute (172x) o3 configuration scored 87.5%.Oh, i thought it was only hitting those numbers when beefed up with investory money, but it's actually the same limit for all models and it STILL is this much of an improvement? i gotta say i was a doubter but holy shit
this also came out yesterdayhttps://www.youtube.com/watch?v=mKSttLFNlW0https://github.com/Genesis-Embodied-AI/Genesis
qrd?
>lose job>get AI waifu to live with in VR>fet UBI so have enough for foodIDK guys I abandoned the thought of being a NEET and became a normie, but this future could make me reconsider
>>103597975>UBIkikes will never allow it
>>103595901>>103596313>>103597052>>103596391>>103595513>>103595480>>103595691>t-two more wee- ACK!
>>103597988Oh you're wrong. Without UBI, people might get angry and rebel to the new feudalism system.But as I said, give me an AI gf and food and I will be a good goy. Kikes love UBI
>>103597993that picture is funny, shows both the retard and also the retards with the downvotes
>>103598004>as long as i only get sodomized once per day it's fine(You) a year in receiving UBI
>>103598018I don't get it anon, please explain.
>>103597993>bud
>>103598055there's going to be strings attached because ZOG, tiny hats view goyim as cattle
>>103598071I know. But it's not like we have a choice. This future was automatically determined once WWII was over, all I'm saying is that in the end it might not be so bad
>>103598162>This future was automatically determined
>>103597993AI-generated images and videos are STILL not photorealistic. He was right.
>>103598204您进入了错误的主题。查看 不和谐 以了解付款详情。
>>103596832It's over.
>>103595476How can you gauge your progress towards a point when nobody knows what that point is?
>>103600283>How can you gauge your progress towards a point when nobody knows what that point is?
>>103595476buy an ad
>>103600700Bot.
>>103595480You obviously don't know what you're speaking on.The very benchmark is impossible to memorize. So try again doubter
>>103600757Goodhart's law
Is this because of a model innovation or huge hardware scaling up?I'm curious if we'll get local models measuring up to Claude 3.5 any time soon. I'm interested in buying a 5090 if so.
>>103600794It's because they're gaming the benchmarks.
>>103600777Mutt's law
>>103600777That doesn't respond to my point at >>103600757cope, seethe and denier harder.Let me break this down to you really elementary like,The test,That o3 was trained upCan NOT be MEMORIZEDWhich meanso3 had to improve its REASONING capabilitiesAnd THINKING capabilitiesAbove that of previous modelsTo reach the percentage it hasIf you do not understand the above, you are literally intentionally being ignorant and trolling
>>103600794Hardware hasn't improved in almost 4 years. They were runninf A100s. Not they have H100s.
>Destroys your illusionCheck out that exponential horizontal cost access LMAOAI is dead
>>103600794The innovation was the o-series, this is them scaling that up. The cycle rn is that they scale whatever architecture they have 'til it hits a wall and then they design a new one and sell us the old one at increasingly cheaper prices.
>>103598168Awwwww. Little cow is getting a head massage form that thingy.
>>103596832This was the attempt to scale up the GPT series and see huge gains. Transformers hit a wall so they pivoted to the "reasoning" models. We'll see how long those new models last.
>>103595476Can someone ask it what th price of Bitcoin will be in a year?
>>103598204are you re3
>>103600881*retardedt. phonefag
>>103600838>it's exponential on an exponential chartit's ogre
>>103600906AI is like a junior team member. he's going to fuck it up, but he can do 90% of the work, mostly the boring shit. if he makes 100k/year, thats like 400 bucks a day. if you can pay AI 100 bucks to get the same result in a few moments, are you still coming out ahead?
>>103601088>but he can do 90% of the workIf you ever worked you would know that last 10% is where 90% of ACTUAL work happens
Until these things can learn and perceive on their own without needing the training step, real brains don't need to be re formed just to learn a few new things.
>>103601108right, thats why experienced people get paid higher salaries
>>103596384This is an English-speaking website.
>>103601088Cope harder
>>103601127That's a dumb take,Babies don't come out if the womb knowing shit.Cope harder
>>103601213go dial your rotary phone on a beach somewhere you fucking boomer
>>103601190You obviously don't speak english if you didn't understand him.Bot
>>103601240>cope seethe shit talking!>I'm mad I can't logically reply to you!!!!!
That graph is cope if I have ever seen it, a benchmark tuned to give their model higher scores and doesn't reflect performance at all.>>103595691This kind of curve is what actual AI performance looks like, the start of the rise is davincii and it plateaus with GPT4 with pretty much no noticeable improvement after that.BTW, I used LLMs a year before you knew what they were. You were busy shilling and losing your saving on crypto back then.
>>103601256I did understand him. It's still ESL-tier writing, and the poster is clearly Indian.
>>103601291You should research how the test isn't capable of being memorized before sounding ignorant in you reply.>>103601292>I understood himSure you did
>>103601309To add to this,This test LITERALLY tests GENERALIZATION OF KNOWLEDGE aka AGIMr. "I only began using AI in 2023"
>>103601309>the test isn't capable of being memorizedEverything is possible to memorize. Memorization is all a LLM can do. Only way to have a test that cannot be memorized is to have a new problem every time. It's still a test that's tuned to give these result because o1 offers at most something like 10% improvement over o4, this is self-evident to any user.Anyway, o1 isn't a advancement in AI, they're just prompting differently. Asking it to go step by step and then prompting to evaluate the prompt. Something we have been doing forever.
>>103601370Ok ok I get what you are doing, you are acting so stupid to make me thing chatGPT is smart. Right?? I hope so.>LITERALLY tests GENERALIZATION OF KNOWLEDGE aka AGIPlease, learn english. These words don't mean what you think they mean. Tell me, what is knowledge and what is intelligence?
>>103601386Again, before you come across as loud and wrong. Research what the test is.Then come back, apologize and stop typing.
>>103601421Lmao you must be trolling
>>103600838>Confucious once said that if you throw gigadzhillion of compute at a problem, you'll get somewhere eventuallyUnless they'll find a quick way to optimize it, ClosedAI are cooked. Being this hardware bound puts it firmly into research tech territory, should've stayed non-profit.
>>103596297>It being trained on the benchmark is not the same as it being trained on the answers.C O P EOPEAt least we now know that the AI isn't the dumbest thing reading this thread.
>>103596701>>103600757>>103600821>The very benchmark is impossible to memorizeactual monkeybrained tardsdo you think the dataset is 100% statistically random numbers? no? then you can train to do it. there is no statistical or mathematical difference between "reasoning" and predicting
>>103601666>>103601713Simple question (and I know the answer is 'No')Have either of you loud opinionated wrong users actually read what the test is and how its taken.
>>103601386>Everything is possible to memorize.No no no. Anon said it was impossible therefore that must be true. Would Anon lie or misrepresent the truth or just be stupid and deluded? Impossible!
>>103601752Waiting for you to post a link.
>>103601752unless the dataset is statistically random information, it doesn't matter
What is the AGI score and how is it calculated?
>>103601810bout tree fiddy
>>103596391This is trueFolks in /lmg/ have known about overtraining on mememarks for a long time, it's interesting seeing the gap of knowledge between /lmg/ and the rest of /g/, with the rest of /g/ being impressed by things like redditors
>>103602056>This is true>Me and another clueless retard agree
>>103595476I'm skeptical because, to me, o1 seems a lot dumber and less 'alive' than 4o does.
>>103601770>Spoon feed me>I can google everything but that which goes directly against my ignorance and narrow minded thinkingIf you truly want to remove your ignorance, you'd google it - I am not going to spoon feed you what is known by those of us who research AI just to have you go;>yea buuuuttttt....Nah>>103601806See>>103601752
>>103601603Last year people said we'd NEVER get anything on the level of GPT-3.5 Turbo because such large models were simply too large and required too much infrastructure.Then more research went into quantization, distillation, and better training techniques and you now can run a model that beats 3.5 Turbo on your cell phone
>>103601713>there is no statistical or mathematical difference between "reasoning" and predictingGlad you finally figured that out. Chinese room tards BTFO
If o3 is so good why didn't OpenAI make it their new CEO?If it can't even beat Sam Altman it's absolute trash.
>>103595476All they need to do to prove it is actual AGI is fire all devs and researchers and just have Sam run the company by typing in whatever he wants to have done.
>>103602558The funny thing is, even those within the CS field who have wrongly predicted slow downs, walls, limits and constraints of AI advancement due to laws have been made to eat crow this entire year.That alone should tell you that doubters in this thread don't carry the weight of an ants fart with their doubts.>Its a scam>Its not AGI>AGI can't happenAll said in this thread by anons who don't even understand AI and there are CS 'experts' who have been proven wrong.Even when ASI is here, these same Anons will seethe hard and try to cope
>>103602660kek this
>>103602684> Even when ASI is here, these same Anons will seethe hard and try to copeThey'll pretend like they never doubted it. That's how it always goes.
>>103602684i like to assume the seethers are highly overrated and overpayed amatuers in the fields of game dev and or the video industry
>>103602660This. And let's say the wagies keep their job. Why would you announce something that is allegedly so amazing that all you have to show is a graph that shows one number if higher than the other? Show a demo, surely for such a profound breakthrough there would be cool demo to show, right? They claim it beats most programmers (based on bullshit competitive programming meme shit) but why can't they show a single example? Or maybe, just maybe OpenAI is doing what they've been doing for a few years now which is generating a bunch of hype for something that ends up being a wet fart.
>>103596297>proves lack of understanding or knowledge in what you speak aboutAnon... I.... You are talking to """people""" that are still trying (and pathetically) failing to argue semantics with the whole hurrdurr not ai bullshit because they still don't even understand the fundamentals about what they are discussing or how ChatGPT, AI, ML and DL are related.
>>103602858Lol damn shame>>103602971Yea, those luddites were pretty pissed earlier this year>>103603860Facts
>>103595476ChatGPT is bullshithttps://link.springer.com/article/10.1007/s10676-024-09775-5
>>103604474>ChatGPT is bullshitOh boy. More philosophy majors who have no idea what the fuck they're talking about thinking they get to have an opinion
>>103604474Based>>103604555Faggot
>>103604474>>103604562Buy an ad
>>103595476At this rate o4 should be released in a week and score 100, then o5 two days later with a score of 300.Oh what's that? None of that will happen because this is bullshit and will be silently memory holed just like every other chart showing how OpenAI is one step away from AGI?They've been doing this shit every couple of months for years now and everyone is sick of it.
>>103596239In the future, employment will be watching advertisements and buying products so advertisers can train better models to sell you more products. They will give you exactly enough "UBI" to continue to live and buying products in exchange for this.
>>103597077>totally useless unless it can do literally everythingBased retard
>>103604724>Humans:>can confidently assert false information>can use flawed logic and reasoning>can fail to perceive information in their environment>can perform many different kinds of tasks with adequate training and examplesBiological General Intelligence!>LLMs>can confidently assert false information>can use flawed logic and reasoning>can fail to perceive information in their environment>can perform many different kinds of tasks with adequate training and examplesTHATS NOT AGI REAL AGI IS SUPERINTELLIGENCE IT SHOULD BE ABLE TO ONE SHOT THE LINUX KERNEL FROM SCRATCH BUGFREE
>>103597077I always had AGI mainly in the ways it is distinct from "narrow" AI. AlphaGO is godlike at GO and only GO. AGI is general because it can apply that same godlike recursive machine intelligence to any domain. That's not what we have and it's never coming.
