>Mythos comes out>it's the same as opus 4.8>which was the same as opus 4.6>which was the same as 4.5>which was marginally better than 4aibros... I don't feel so good
>>109102323>I randomly decided to put everything on a straight line to pretend nothing’s changing cope
>>109102323>low iq retard used MMLU, a basic human benchmark that models maxxed out by now, that also contains a lot of errors btw, as a cap for all model capability forever>retard is then surprised that the benchmark, that is surpassed by now since a lot of models clear it completely, shows that all new models... clear it completelykek, shows the average anti-ai faggot is literally braindead.
>>109102323Mythos was just 4.8 with ultra max megathink
>>109102323Luddite cope thread
>>109102323>WHY WON'T IT GO OVER 1 WHAT IS HAPPENING
>>109102395I really wonder how much do you get paid for posting pro AI shit on every thread.
>>109102442Your seethe is my payment
>>109102323shouldn't those modern MPT, E, and polarized coord models have show a decent jump?Zai glm 5.2 as well
>>109102323https://archive.ph/aA1dB>On June 11th Mark Warner, the vice-chair of the Senate Intelligence Committee, said that General Joshua Rudd, who leads the National Security Agency and the Pentagon’s Cyber Command, had told him that Mythos “broke into almost all of our classified systems, not in weeks, but in hours”.a us senator said mythos broke into the nsa's most protected systems
>SOME DUMBASS ON TWITTERNOSTOP REPLYING TO THIS GARBAGEYou need an ACTUAL SOURCE if you want to claim something. Some idiot on twitter posting a graph and pretending like it's credible research does not meet that bar. Grow a brain.
>>109102570Mark Warner is a 71 year old lawyer and fund manager, he's ricing the IPO or doesn't know how computers work.
>>109102395good morning saar
>>109102655You’re trans
>>109102570this makes no sense operationally, did this happen in a wargame scenario or we were attacked by someone?how do they know that its the model and not just the prompting/contextsee if this is wargames I'd find it important to know if the model has the same guardrails it normally does, and secondly if they did prompt it to find a novel solution to a crypto algo, I don't think the AI is at fault. like this really sounds like mythos found a zero day in some system after analyzing code and now some faggot boomer is trying to paint that as hacking all our systems
>>109102442Base pay 0.01 USD per post, fresh gens net him 0.20 USD first time posted, per-post bonus of 0.02 USD for using term "luddite" or "luddites". Thielbux are sent at the end of the month and he'll be able to buy a couple cans of coke with it.
>>109102777You're queer!ew! (vibequeer!)
>>109102442how much do the voices in your head pay you for being obscenely paranoid and delusional that multiple people don't have the same opinion
>>109102795>AAAAAAAIIIIIIIIIIEEEEEE!!https://desuarchive.org/g/sea!rch/image/x0gKNuxAY6jzo5YKKoKtEg/https://desuarchive.org/g/search/image/cjZAjIJm-irBTz-yLrJB7w/https://desuarchive.org/g/search/image/C_iyRk39MpkDGJNZUKN-Pg/
>>109102795Dude, its tiring to have you posting on every thread every single day. Its not even subtitle. I have news for you, AI didnt close the breach between software engineers and normie retards like you at all. All AI did was to make SaaS obsolete as a business model. But any former pre-AI engineer will be way more suitable to use AI than any normie just because they can make more accurate request to the slop bot.
>>109102323It wasn't the same. It felt like I was talking to an alien intelligence. If you had tried it the 1 day we had to use it instead of gooning you would know
>>109102819>Dude, its tiring to have you posting on every thread every single day. Its not even subtitle.you using AI to generate this post isn't subtle eitherI just cannot believe someone is this retarded
>>109103228Christ. How fucking dumb do you have to be to think you're talking to intelligence when chatting with an LLM
>Mythos comes out>it's the same as opus 4.8Definitely not. It was noticeably better when I tested it before they took it down. Benchmarks also show significant jump.
>>109103257>Benchmarks also show significant jump.Which are pretty retarded since you can overfit for them, so people have to come up with new ones all the time so that they aren't included in the training data
>>109103264and mythos was better on all the new ones toomodel capability now is at a point where assessing them has become a test of the user's intelligenceif you can't tell that mythos is better than opus 4.x then you're dumber than opus 4.x
>>109103272>and mythos was better on all the new ones tooand why do you think they aren't part of the training data? fine tuning is not that hard
>>109103244Intelligence is the ability to solve non-trivial problems. Fable definitely did it for me when I was using it (solved some hard bugs, improved the design of my project). This means LLMs can be intelligent.
>>109103277you're going to be saying this as claude 10 cures your mother's cancer
>>109103277Many benchmarks keep private test sets.
>samefagging in this thread is off the charts
Luddites getting BTFO ITT.
>>109103290the models they use in spotting cancer don't use LLMs tho>>109103295jeets are gonna jeet
>>109103306>spottinglol. at this rate claude's going to have to cure the cancer i'm getting from reading your retarded posts
I don't understand why it's called "AI". Intelligence implies some sort of persistence and some sort of processing/integration of experience, which can't really happen with a stateless statistical machine.
>>109103312ok?
>>109102570That dude will probably make a lot of money with anthropic IPO so he is just hyping it up for max profit. >trust me bro
Luddites are delulu
Clawmoltmasturbationshills ITT
>>109103332>Intelligence implies some sort of persistenceJust use editable `memory.md` and append it to context with every prompt.
The most laughable part about all these pro-AI shills is that they aren't even paid for it. They literally do it for free.
>>109102323>release a new model>granularly reduce the available compute for your older models>claim now that the new model is 10x in comparison to its predecessor model>rinse and repeat, claiming exponential improvements, while in reality they are way below lineartired of this trend. you can clearly see this happening with the current opus.
>>109103400That's not really what I meant. I meant something like the hippocampus, integrating new memories right into the model.
>>109103332"intelligence" is just a metaphor. just like "object" in object-oriented programming.
>>109103463i think it's like a couple of very mentally ill guys who developed para-social relationships with their bots. They wish they were paid lol
>>109103514It's not a metaphor, it is a concept (maybe that's what you meant). Intelligence is what make us adapt and to adapt you must constantly learn from your environment and your experiences.AI doesn't have that, it's not intelligence, it's just a big statistical machine. I don't judge on its efficiency, because obviously LM are a great concept that can do a lot, but I'm uncomfortable calling it "intelligence".
>>109103530AI can learn to some extent. In-context learning is still a form of learning. It is limited compared to actually changing the neural network weights, but it is a form of intelligence and adaptability.
>>109103530intelligence in ai is used very broadly. you can call a couple of if statements as ai in videogames. Anything that mimics some part of reasoning can be called intelligence.for llmslop companies they obviously try to bamboozle people by applying this term as full human intelligence replacement.
>>109103284R-r-r-results?
>>109102333>RandomlyOh no, it's brain damaged and shits outside.
>>109102442He gets paid with attention, which he wouldn't get otherwise.
>>109102795Nobody's paying this needy, deranged fuckup, he does it for attention. He thinks he's doing le "ebig trole."
>>109103576That line graph claims the old gpt-4o has the same level of capability as gpt-5.5. I don't know what statistical fuckery he did to obtain it, but it is clearly complete bullshit. Just try to vibecode something with gpt-4o (or available equivalent) and gpt-5.5, then compare the results.
>>109103463a lot of them are paid nigga and this shithole is just one of the many places they shillyes there are those who do it for free but a fuckton of AI shill threads are unironically posted during Indian morning hours
>>109102367> mmlu was already a joke benchmark a year ago and they're still using it as a gotchalmao imagine being this late to the party
>>109102793> 0.01 per post and a bonus for saying "luddite" is some real hustle culture cope. enjoy your two cokes, king.
>>109103605>but it is clearly complete bullshit.Prove it.
>>109103620Yeah all those generated macros aren't being prompted by one idiot. This is getting muscovite attention.
>>109103677in retrospect it's not surprising it's being shilled in countless places. There are tens of billions being thrown away each year on marketing and main userbase of these systems are nerds who might browse this shithole.
>>109103669>prove that gpt-4o and gpt-5.5 do not have the same level of capability, as my retarded graph says!>oh, you won't spoonfeed me basic benchmarks? Guess I win!This is the level of argumentation luddites seriously consider persuasive, lol. No, I won't spoonfeed you. Have fun coping and being left behind.
>>109103705oh look a saar>5 pm in indiaworking overtime?
>>109103647Are the luddites in the room with us now?
>>109103735AI is peak White and Asian technology. Keep coping, neo-Luddite. You will end up exactly like your historical counterparts.
>>109103760Everyone is using AI, especially Indians.
>>109102323>some older benchmarks saturateduhnor really a fan of ai but OP is a faggot
>>109103763the more AI the less reasons to import indians
>>109102323the meme has to continue for a little longer, our economy depends on ittrust us 2 more weeks until ubi
>>109103768benchmarks are outdated the moment they're posted. Everyone games them around them. You're a faggot too.
>>109103778ai using indians are cheaper than you. Their main usacase is to take the blame anyway. Sorry nothing changed buddy
>>109103789it will continue until early investors cash out and socialize the losses not until ubi lol
>>109103799>trying to make a poor argument in favor of literal who xitter screencapthere are better reasons to hate on aiffs you are better than this
>>109103705>Have fun coping and being left behind.I can't be left behind because I'm not some bootcamp faggot doing webdevai can't do what I do and it's very likely (though not certain) that the ceiling will be bellow what I can doif you're impressed by what ai can do and produly claim that you never write code anymore, that's a good sign that you will be easily replaced. you could be replaced already, it's all a matter of the tooling catching up, if it hasn't already
>>109103806you called out the op on benchmark being too saturated, i call you out on thinking benchmarks matter at all. Not rocket science buddy. I don't agree with op that exact tweet matters either
>>109103813i didn't say benchmarks can be absolutely trustedbut at some point, you have to anchor your logic to measurable evidence if you are making arguments for something in a quantifiable mannerif you think benchmarks are purely meaningless numbers without any trace of truth, idk what to say to be honest
>redditor uses a saturated benchmark with an ai generated graph to prove ai is hitting a wall>thinks he is very smart
>>109103851>if you think benchmarks are purely meaningless numbers without any trace of truth, idk what to say to be honestfor the most part. They're not reliable for measuring which model is better. so yeah le number bigger doesn't mean shit.>can they measure SOMETHING?probably lol, maybe the first time they're posted. The moment it becomes a standard they're gamed around easily.
>>109103605So why can't they patch this bug for a year? (https://github.com/anthropics/claude-code/issues/769) The closest they have came to patching this so far is an environment variable that disables basic functionality and "works 85%" of the time.
>>109103809>if you're impressed by what ai can do and produly claim that you never write code anymore, that's a good sign that you will be easily replacedbullshit, the opposite is true. I regularly chew through my claude max tokens but Claude could not do what it does without me, and my role could not and will not be replaced by some better model.You luddite code-writing fetishists are cute, but you're not irreplaceable.
>>109103906>and my role could not and will not be replaced by some better model.cute ai bro cope, you will be replaced even faster lol
>>109103902HAHAHAHA, they really do work with vibes
>>109103906In six months some business major will just ask Claude to take the transcript of his meetings with the client, generate a bunch of tickets (for stakeholder visibility, mostly), and then go get a coffee while agents implement, test, and deploy the finished product. I’m sure that could be done already by some more resourceful non-tech people, but there’s still some friction in places. That will go away soon. You are not necessary
Anthropic updated its Privacy Policy this week. Starting July 8, 2026, certain capabilities will require identity verification and it will be handled by Persona.Persona is a third party identity verification company backed by Peter Thiel. The announcement comes months after Discord dropped Persona as an age verification vendor following significant user backlash and a reported data exposure incident in February 2026.
>>109104011The self-proclaimed VibeGOD shills don't have a response to this. Oversight of the agents ("agent managers") will be done for <$1/h by Filipinos. The European VibeGODs will join the underclass with us and their good boy tokens will be rescinded.
>>109104011>while agents implement, test, and deploy the finished product.i can already imagine the future>clause pls implement this shitty crud app>that will be 500 million dollars plus tip, thank you!>uhhh pls fix the weird flicker problem? no mistakes, run 100 agents> mistake fixed, another 100 million + tip>it still flickers bro.....
>>109104035KEK there's an official start date now? The policies were changed a few weeks ago to allow this and pave the way
>>109102452Nobody's actually seething though. It's all in your head.
>>109103579>$0.10 for every (you)Yea, but who's paying it? Sam Altman?
>>109103705>ludditesI spend more time on the computer than you.
>>109102395AI is a bubble, an internet like one where all the retarded shit will be extincted.>>109102323Bruteforce fags will get mogged.The future comes from the architecturefags, that will create complex AI machines with many specialized parts that include llms, neural networks and hard code.As soon they gave a python prompt to grok, it was all over to "JUST ADD MORE POWER BRO".The AI of the future will be a brain.If we don't literally use actual artificial grown brains in a jar.
>>109102323Its called testing limitation. If you cant recognize 100IQ from 110IQ from 150IQ from 200IQ, then the problem isnt the person you're testing, the problem is your own testing limits
>>109102395What the banana mean
>>109102570Yeah I was saying you could just prompt it to hack the US government but I dont think people believed me. tbf I had no source it was a guess.
>>109103463>>109103518They're not paid, they're eternal losers who failed to ever learn something because they gave up immediately, so they see AI as both their ticket to success and a way to get back at people who did have the mindset to learn and succeed. Just go look at /agdg/ over at /vg/: the entire thread collapsed because nodevs, after years of seething, jumped on the AI bandwagon thinking they'd finally make it. Instead they're stuck vibeslopping shit that doesn't work and project hopping just like before which just makes them more bitter and entrenched.
>>109102333Imo it’s a bigger folly to try and act like you can measure intelligence.We can’t even decide if IQ is truly good at measuring intelligence.Don’t think they’re proving much with any of these Mctests.
>>109104035>>109104046Grifter Theo is still trying to inject himself into the discussionhttps://x.com/theo/status/2068673461061550262>NOOOOOOOOO ITS A HECKIN MISINFORMATIONERINO!!!!11!
>>109103305
>>109103284Solved some bugs? Sure, it has the entirety of SO in its "memory" and frankly your bugs aren't likely to be uniqueImproved the design of your project? If so then it had to be worse than abysmal in the first placeAI is effectively a retarded savant, with virtually no reasoning ability but really good memory and pattern recognition
>>109105044i wish it solved bugs as well as ai fags claim it does. It works on some trivial stuff, but when you hit some more advanced shit it's utterly useless. Funnily enough it works even worse on its own vibeslop. Probably not enough vibeslop repos in its training data yet, ironically. Or maybe they're trying to avoid the inbred version of model trained on its own slop.
I know the obvious answer is that you (or they) are dumb and drinking their own koolaidBut why exactly do people think that some new magic emergent behavior will spawn out of throwing more compute at the same AI models rebranded as new ones when they've basically scraped everything and there's no new data to steal from?
>>109102792It's probably misleading. The government had it for testing so they probably gave it a way to access pentagon and NSA systems and see if it could penetrate defenses once it was on the network. The US always tests in worst case scenarios. Wargames are a good example of that. It's common for the US military to test worst case scenarios where they're not able to use any intelligence, or like half the equipment they have, in order to test the floor of their capabilities.
>>109105107the era where you scrape, train off of that and call it a day is long gone thoughi am not saying it is not problematic but unironically throwing more compute can in fact create somewhat useful data to train back
>>109102323Incorrect. AI is exponential. We are at the knee of the exponential and entering the Singularity.
>>109105162i dont hate ai but people like you make me sick of it
>>109105152synthetic data still needs someone to verify it tho it's not just compute but also infinity jeets working on it.
>>109105191you can synthesize on verifiable mattersfor math, stuff like lean, unit tests, compiler constraints etc..
>>109105198if the verifier does not cover, it goes elsewhere but they are doing pretty much everything imaginable at this point
>>109105198it's cool and all for specializing models, but this shit isn't producing agi by itself probably lol
>>109105255no shit sherlockthough it still is kind of cool imo
>>109103230isn't that exactly what you want though?
>>109102395>will be executeddon't need to retell your plans, mason
>>109103230AI wouldn’t make such basic grammar mistakes and AI is retarded.
>>109102323if that were true every AI company would IPO before investors figure what going on
>>109105699guess what's happening this year lol
>>109102792You are getting reports through two layers of illiterate boomers, at least one of whom has a vested interest in pumping his bags
>>109105744they really are going to give this exploding bomb to a bunch of retarded boomers and just walk away with it after inflating this retarded balloon for four yearsthey are gonna get away with it like in 08 and just move on to the next griftnothing will happen to them maybe altman will be in court for a bit but thats itmaybe the goyim really are cattle
>>109102442You responding increased his pay. Don’t engage
>>109102323SNAILCATS WIN AGAIN!fight me
>>109103463the shilling on here is absolutely paid for. a few are doing it to troll because its so dumb and annoying. but by far its a campaign to draw people to particular models and get people using it.literally half this board is astroturfing at this point
>>109106297yep. most likely.Although depends if this whole shit ends up crashing the economy, since almost entire us gdp growth last years was due to ai. if it gets bad enough they might throw one of those dipshits to the dogs.
>>109102323I have been saying this for a while and nobody believes it. Since GPT 4 all the perceived improvements have come from overfitting what people actually use LLMs for (using gathered usage data), not from actual "intelligence" of the LLM increasing.
>>109103503I'm okay with this. GPT 5.5 helps me a lot with writing simple code fast, reviewing, etc. "AI" can stay where it is. Just make it cheaper and come up with strategies to make running actually useful models locally possible.
>>109103463>>109106344the snailcat/buffcat bullshit was 100% some thinktank inventionyes, people carry the torch because they think its funny. but they didn't come up with it
>>109106475i used to think it's retarded, but there are actual real think tanks nowadays where retards engineer new memes to public for propaganda purposes
>>109106475i suspected it was just one retarded jeet marketing team doing that. thinktanks are absolutely capable of developing memes that aren't purely retarded.for example the amerimutt meme was definitely one of those. i watched it spawn in one day on pol and people had 100s of OC ready to drop instantly. literally 0% chance it was organic and it was very effective
>>109106511i have no idea what kind of 4d chess they are trying to doit is way too atrocious for 'pro-ai'and those so called thinktank gain what from crafting 'anti-ai' meme in disguise of pro ai on purpose?i dont get it
>>109102323don't listen to the luddite propaganda, boys!
>>109106549probably bait to attract more attention and discussion. These threads last longer.Anti-ai doomerism angle (agi will kill us etc) is also good for them. Makes you think their autocomplete bots have something more to them.
>>109106573so a part of 'number go up regardless of context' campaign?i dont personally hate ai but seeing those kinds of stuff sickens me
>>109102367>>109103629>uhh dis benchmark no counti rememeber when goymark fags were posting this same shit about meta's new AI>OH MY SCIENCE THE BENCHMARKERINO AGI IS HEREhow'd that go, faggot retards?
>>109106591i think so. it probably got much worse recently due to ipo's happening this year. Everyone wants to max their value and dump it on boomer pension funds asap. They're using any tactic possible for marketing
holy botted thread
>>109102323But the b00nchmarx....!
>it's the same as opus 4.8With the slight exception that Mythos is literally classified as a potential Weapon of Mass Destruction and Opus 4.8 is not.Yep, basically the same. /s
>>109107635>a word predictor is a weapon of mass destructionai psychosis induced retardationyou hate to see it
>>109107686The US gov't literally had to shut it down.>word predictorThe same can be said of people.
>>109107688>animals capable of thoughts are the same thing as weights stored in a disklet me guess, cats and dogs are also "word predictors"?
>>109107702In a way, yes, except they don't use words. But tokens and symbols? Yes, absolutely.
>halving the error rate with each generation is not exponential improvementRetards really see Model A with 0.98 accuracy and Model B with 0.99 accuracy and think Model B it's only 1% better.
>>109102323Claude is fucking shit now lol good riddance
>>109102799Obsessed and mindbroken
>>109106598and then the shapoo bottles clapped
>>109104487its a reference to theamazingatheist
>>109108344He trooned out? Haven't checked in on that particular lolcow in a long while.
>>109108146Spammed the same images on a public imageboard award
Humans cannot reliably evaluate what is superhuman, therefore any further increase in capability that transcends the ceiling of human capability will be perceived as a flat curve.
>People sincerely arguing the benchmark should be over one thus progress has stalled>People thinking the benchmark is fake because the values don't go over oneIncredible, bravo. Genuine all pervasive next level retardation from /g/ once again
>>109102323In what fucking world is gpt-4 as good as mythos
>>109106297Better yet, SpaceX successfully lobbied for relaxing or removing the regulations for freshly IPO'd stocks being included in retirement fundsSo not only are they about to tank the economy, they'll take everyone's retirement with them (which is of course why they'll be declared "too big to fail" and bailed out on taxpayer's dime)I fucking hate communists like you wouldn't believe, but I'm starting to think they might've had a point with the "eat the rich" thing because goddamn
>>109103579good point, it seems like every general on this site has a dedicated attentionwhore schizo who likes to ruin threads just for the love of the game
"predicted"herbs.
>>109109083>Humans cannot reliably evaluate what is superhuman, therefore any further increase in capability that transcends the ceiling of human capability will be perceived as a flat curve.This is the dumbest take i have ever read. You are the blackest, niggerest gorilla nigger in the hood.
>>109102323I also made a graph showing how AI scaling is extremely exponential. It is and still is going up __ AlexNet GPT 543. R56 = 9.98. The Ass Symbiote is infinity entering this guys anus cavity in and out.It was and will be. /THREAD
>>109110626You?
>>109109801Eat the rich has nothing to do with communism and everything to do with being and "anticorporation bootlicking libertarian".Common ways to figure out of someone is a fake libertarian:Ask them to include tax in prices across every state. Instead of defending consumer rights, they will say "it takes 2 seconds to calculate sales tax" instead of pushing blame/burden onto corporations saying that it's "too difficult" to implement sales tax for different states logistically (it isn't).Too lazy to think of anything else but you get my point
>>109110703AGI is not superhuman, you retarded nigger. ASI is. A model to be considered AGI just has to UNIVERSALLY perform at a human level.
>>109102323wrong. stacking enough GPUs inside a dyson sphere will give us AGI
>>109110926
>>109102442You’re on the technology board retard why the fuck are you even here if you hate AI? Kill yourself and save us the trouble.
>>109102323ai is already good enough to cum to. what more do we need. i vibe coded a project to when im roleplaying with my agent it will generate images with comfy ui of what she is doing and post them with the responses
>>109102370VibeGODS have no response for this one
>>109102452KEK based.
>>109106297>they really are going to give this exploding bomb to a bunch of retarded boomers and just walk away with it after inflating this retarded balloon for four yearsThey deserve it
>>109111762>yeah bro if you don't love every single bullshit that comes from silicon valley you hate technology broyou're a real fag aren't u
>>109111762>you must love ALL technology
