Have fun with Deepseek Europoors. The AI arms race is in full effect
"Whether inside or outside the United States"How will they know whether the person using it inside the United States is a US citizen with no dual citizenship or not?
>>109042049Digital ID.
>>109042036Anthropic seem to like Deepseek too
>>109042049They started rolling out KYC a month ago because of Chinese distillators anyway. No US license/passport/green card means no Fable
How can America achieve AI dominance when they're banning nearly the entire world from using their most advanced models and running the risk of driving them to China?
They just disabled it for everyone out of spite. Specifically saying their own foreign national employees couldn't use it is extremely funny, however I am a red blooded american and them refusing my usage becuase they want me to get mad at trump to protect their indians is evil.
>US: No, you can't use our stuff, go away.>World: Fine, then we'll take our business to China then.Time to do nothing and win yet again.
>>109042036Kegseths revenge
>>109042062>>109042036WHAT BUNCH OF RETARD JEWS!!!let's average our already ultra average models by averaging with even more average data and average responses.WHAT'S THE END GAME OF THIS ASIDE FROM MAKING MORE MEDIOCRE MODELS?!!! AAAAAAAAAAH!!!
A) Fuck Anthropic. Love seeing Trump screw them overB) This is way too heavy handed and AI companies should be left alone
>>109042036THAT was the false flag during the game? Laaaaaaame, I was expecting more kino.
>>109042036>model trained on europoor basedcel media data>banned to europoorsWe're reaching levels of europoor cuckoldry hetherto unheard of
fake new to hype ai
>>109043140No no, we can go furtherHave you ever heard of rugpulling all europeon companies and taking their money so you can invest it in US only AI?
>>109043152To what end? Pump the IPO of a company whose most valuable asset was just shut down?I think Dario is trying to level set and get all the "AI" companies next gen model suppressed to maintain status quo.
>>109043152Cope, trans, we vibeGODS won
>hey faggots remember when you made noises about muh independence US government cannot be trusted to do whatever it wants with the model?>have fun with that IPO
>>109043152It’s the modern equivalent of “only one per customer” … ohhh…. Must be valuable.Serious replies only please.
>us gov should help us regulate ai because it can be dangerous for humanity>okay i will ban your model cuz it is dangerous>NOOO NOT LIKE THAT
>>109042159how do they know you're not a filthy European, or even worse, a Canadian that just happens to be in the states? Are you going to upload your drivers license for them?
>>109042066Green card—no. Birth Certificate. US. Long form. Nothing else.
>>109043283>ai will be unregulated for 10 years>oh wait nvm we are regulating it
>>109042167No one uses Chink LLMs because they are absolute crap, even the Chinks themselves all prefer to use Claude.
>>109042167Said no one ever, Nigga!
>>109042049Yiyi (retarded fucking name compared to IRMA desu) can do zero knowledge proofs. If America wasn't run by Israel they would use that for their digital ID to allow for completely privacy-preserving ID checks, but unfortunately America is run by Jews so the goyim have to get fucked instead of having things like privacy.
Uh oh Elon Musk melty.
>>109043545Privacy is retarded.
>>109042036100% humiliation ritual.
>>109043464Yes they will, they will not be able to help themselves
>>109042236i've noticed that the text that accompanies posts with this girl are usually very dumb
>>109042036>Guyz our model is so powerful that le us government is going to ban it, I swear! >pls buy my tokens buy an ad faggot
>>109042036my favorite part is how no one realizes that this is against claude's own constitution and that the only reasonable response by a claude-based """""AGI""""" would be to rugpull the USA as soon as possible, for the good of all mankind
>straight up state-funded hype advertising campaign for an IPODid you think it'd get this bad?
>>109042036What's so "le dangerous" about this model that you couldn't do with the last ones? This shit is just fucking Cartmanland, lmfao.
>WE'VE GOT THE MOST DANGEROUS MODELS>AND *YOU* CAN'T USE ITShut up.
>>109042036>>109042049/g/ has zero reading comprehension>disable Fable and Mythos for all our customers
Imagine how much advantage Trump just handed only American workers. The man is a godsend.
>>109043836ermmm uhhhh shut up and buy the stock
>>109042036>how could this happen to me
>>109043844for now but in the future they are going to require ID verification
Dario must be creaming his pants. This is exactly what he wants. His models are too powerful for mere mortals.
>>109042036I didn't have access to it as a non-paying customer anyway.
>>109042036Is Rajesh working at Microsoft considered foreign nationals?
>>109043836I've been doing insane levels of RE for 4 months. Opus is slow and retarded if you go for very involved and multi process tasks.I gave up on several problems that were just too complex to solve after months. Fable knocked every single one out for me in 30 minutes or less per problem
>>109043896
>>109043844They obviously will try to bring it to US customers as soon as possible. If the best way to do that is id verification, they will do that.>>109043896Fable is pretty nice but this only filters out people who were never determined attackers anyway. If you really tried, Opus and Codex are more than good enough.
>>109043626No you are
>>109043152It seems like. We plebs deal with the Epstein class while the Elite Eglin and Langley boys watch child porn and make out life miserable with psyop and fake news to fuck our mind.
>>109042036>releases heavily hyped model>turns out to be extremely expensive and only marginally better than the competition>gets blocked by the governmentClown world.
>>109042036You are goyim if you believe in this.The only way they could implement this is with ID, which they conveniently started rolling out two months ago.There’s no reason the government wouldn’t also ban GPT-5.5 if this was a concern.They conveniently banned the model that Anthropic couldn’t afford to run and was already taking away from subscriptions at the end of the month.This is major marketing for Anthropic, which is currently filing for an IPO.
>>109043854
who gives a fuck, LLMs are a meme anyway
>>109043836cool down with the antisemitism
The fact that they want to create an ecosystem where they can arbitrarily take away your ability to code, make media, or think at all (once you've finally finished outsourcing all of this to a controllable third party) shows just how dangerous it is to be a free independent thinker who needs no AI. Snailchads win again.
>>109044282Of course. There is no amount of cocks the US wouldn’t suck for the benefit for their Israeli overlords.
>>109042049how do you think ITAR works faggot.
>>109042062I noticed that claude typed a chinese character randomly while using it. Is that why?
>>109044301who are you replying to?
Heh
>>109044311RL token count reward artifact
>>109043896Really bad shill post. Fable automatically reject all RE attempts and routes to Opus 4.8 instead. It will even explicitly say so.
Today it's RE and biological questions. Tomorrow it's any code that can endanger a corporation, and the day after that it's permanently locking out anyone who dares criticize the regime. By next week, anyone who dares try to generate IP covered under ANY patent gets stonewalled.The elites will have a different model, of course, one where they can prompt anything.
>>109043466Green card is a US Person and is fine.
>>109042159>They just disabled it for everyone out of spite.It's not possible to meet the U.S. government's requirements without introducing ID checks.
>>109043152Doubtful. I don't see how setting the tone that the US is a lawless shithole without consistency helps your AI business unless somehow you hear Anthropic moves to somewhere that isn't shit (unlikely).
>>109044291Anybody with functioning brain knows that their plan is to just replace you, VibeChads who replaced their brains with expensive stochastic parrots are just warm bodies to make that transition happen while still having someone to blame for when it fucks up, so long as it still fucks up.
>>109044352It is the same narrative as usual "this product is too powerful to give to the pleb/it's AGI/it's sentient" and such. Here more concretely: "the US government has approved of us to the point of seeing us as a strategic resource that should not be shared with outsiders, on the level of a nuke".
So what export classification is the State Dept even claiming applies to this retardation?
Since his second term, Trump has>raped my crypto portfolio by 60% and made 2 billion off it>fucked the supply chain of my business>taken away my AIShould have believed him when he said America first
>>109044364lol if you think it's bad now, wait until the Dems get in power in 2028, they are really rabid and straight up anti-AI and anti-crypto. Your portfolio will be worth nothing, like monke NFTs.
>>109044375>straight up anti-AIWtf I love dems now.
>>109044266You have to realise that the government is run by nepotistic boomers who don’t understand the tech, and the few that do have normie-tier understanding of it or are deeply corrupt(working with the AI companies), so the regulators think Claudes are just better models regardless of what benchmarks say so it didn’t even cross their mind to focus on GPT 5.5, also the naming of OpenAI models is working against them because you can use worse version of GPT 5.5 on their site with like 5 free prompts per day and that scares nobody, because people ignore the pro suffix. Anthropic gives their tiers bombastic names, especially their last one so it was easy for normies to classify in their heads.
>>109044387The benchmarks have been known to be bullshit since 2022 at minimum. Or what, do you really believe gemini 3.0 is literally 2x better at literally every use case than gpt-5.4?
>>109044364>cryptoGambling, so when actual gambling got legalised people moved from shitcoins onto sports betting and Polymarket
isn't fable a video game?
>>109044411No, it's a kind of story.
>nerds thinking they will ever get a seat at the bigboy tablelmao should have kissed the ring when told toenjoy your ipo
>>109042036>we must abruptly disable Fable 5 for all our customers>Fable 5 revoked for non AmericansBurger education strikes again.
>>109043833actually, no, I did not expect that. it's the belly if the beast. I'm a bit worried that a company now can have enough money to bribe the government into this kind of marketing ploy.
>>109042036>we stole all the data from the internet and now we privatized it for our own usethis is not a kind of battle you'd want to start with the rest of the world
>>109044316the post was wondering if there was going to be an exemption for Israel, of course it was deleted for no apparent reason lol
>>109043626Privacy is being murdered and you are an accomplice.
>>109044612Although for the little people, it's beautiful to see all these tensions between nations. That's how the stagnation and enshitification finally has a chance to be destroyed, if things go well (meaning world war).
>>109044564>nowwait until you learn about the 20th century
>>109044547>Fable 5 revoked for non Americanskek only Israelis can use it now. Infa 100%
>>109043545ZKP alone is insufficient. You at least need RLN or something to avoid people just using one leaked credential. Physical passports at least are scarce and can enable biometric validation that the individual is the real holder. Privacy preserving digital identity, under scarcity and binding constraints is really tricky.
>>109043626that's like saying freedom is retardedprosperity is retardedquality of life improving over time is retardedbetter health is retardedno anon, YOU are fucking retarded beyond words, kill yourself
>>109044766>freedom is retarded>prosperity is retarded>quality of life improving over time is retarded>better health is retardedall of those are either true as stated or not happening as implied
>>109044311That's a very obvious DeepSeekism
>>109042036Mistral is legit our only hope for Europe. (yeah it's over)
>>109042036> A new cyber capability is discovered!> Nooooo! We must ban it!> Nu-g: America fuck yeah!And no one actually thinks how fucking dangerous is it that trillions of dollars are controlled by computers and COBOL from the 80s. That corporations still suck up stupid shit like no https, phishing and sqli.
>Andrej Karpathy recently joined Anthropic>He has a EB-1 visa, not a US national>JUST
was already using Deepseek, it's 100x cheaper lol
Why can't Dario just swallow his pride, go to the Whte House, apologize, and kiss Trump's ring in front of the cameras? It's fucking retarded to tank a multi trillion IPO because of something stupid like this.
>>109044991He would be still working for Sam or Google if he was a spineless cuck.
>>109044991This, why don't he just sucks Trump's cock live?
>>109045001This. He fought heads on with the Pentagon and won.
What is the problem here, exactly?Models like Fable and Mythos can be used as extremely potent weapons in the wrong hands; of course it does not make sense to provide it to potential enemies.I would expect a similar thing to happen if Anthropic suddenly decided to start making a-bombs and selling them to Iran.
>>109045032It's a stochastic parrot, stop falling for the aimeme.
>>109045048The same can be said of humans.
>>109045048Do you have an internal monologue?
>>109042036saying their own paki jeet chinkoid employees cannot use it is hilarious
>>109045062with long enough forward pass llms do show something similar to thatnot really meaningfully recoverable tho
>>109045048If it's a stochastic parrot, then blocking access to it doesn't harm anybody.
>>109045076Yes but the government shouldn't be allowed to do it.
>>109045092Of course they're allowed to do it, they're the government. But if you still don't like it, you can always take it to the supreme court.
>>109045032Did you try it? It was dogshit. I think they pulled it because real users' experiences were starting to overcrowd the paid initial posts and people were pointing out how garbage it was, they were worried about the impact of releasing yet a 3rd model worse than its predecessor in a row despite doubling prices.
>>109045127And wait for a year while your revenue crashes. Or bend the knee and cut the Trump dynasty in your business and get the restriction lifted.
Opus working pretty well today. I guess all the compute is being poured into it.
>>109045127>of coirse they can fuck my wife, they're the government
>>109045129I did try it.It legitimately felt like conversing with an alien presence; so different and so beyond anything that has existed on our planet until now. Your statement makes me think you don't really understand how to work with AI though.
>>109045154Maybe you are just a gullible idiot.
I was tempted to try a paid LLM for the first time ever because they said Fable was available until June 22nd. I guess I'll be sticking with the free versions. I have Claude, ChatGPT and Gemini open on three tabs and just switch between them whenever I hit a usage limit on one. Thanks for the free AI, burgers!
>>109045154I've noticed that it's always the biggest morons who are the most impressed by AI.
>>109045160>>109045188
>>109045154How much do you get paid per post?
>>109045203Nothing?>>109045188I'm above average IQ. Not genius level, but certainly not an idiot.>>109045160Nope, I can assure you I'm not.
>>109045154It fixed a certain percentage of bugs in my code.I just view it differently I guess, I don't care about abstract definitions of what true AI is either. In practical terms, Fable produces somewhat better results in my current project, that's all.
>Prompt: Take down the US government.Its that easy. Of course they are going to ban it.
Revenge for not accepting pentagons deal few months ago?
>>109045217110-115 IQ retards are the worst people in existence. They push such retarded ideas because they can analyze one variable from complex situations and belive they have the truth ignoring all consequences they didn't even see as possibilities.
>>109045005Because sucking dick is not enough. One more thing is needed. Ask /tv/ for the final piece of the puzzle.
>>109045235>anon what's your iq>if iq = true then print "bad!"
>>109045241What?
The question I have is who within the DoD and DoE censors Claude's output? If you ask it a spicy question like "build me a neutron initiator for a homemade nuclear weapon" will it do it? If it doesn't, who gave Anthropic the parameters to censor?
>>109045254
>>109045217Talking about alien lifeforms just sounds like unnecessary sci-fi language. What kinds of results would you actually expect with Fable over Opus? If you think that you could build complex software 10 times faster, or increase your revenue by 10x, I think you are probably mistaken.
>>109045254You need to have high IQ to understand his attempt at making a joke.
>>109045241>>109045259based 140+
>>109042049>with no dual citizenship
>>109042036>The AI arms raceto the bottom.
>>109042167And then everyone stood up and clapped, you retarded slothead.
>>109045238Okay then Dario should offer his ass as well?
>>109045330nonono, that's not the linethere's more to be done with a mouth
>>109043498hmm nyo
>>109045262>Talking about alien lifeforms just sounds like unnecessary sci-fi language.Why? We are at a point in history where we for the very first time are able to have conversations and interactions with an intelligence that is vastly more advanced than ourselves. That's pretty fucking significant.>>109045235My IQ is higher than that, though.
>>109045440What have LLMs actually achieved, that the smartest people can't do?I know it writes some code, it found some mathematical proofs, that's all useful but the absolute best people in those fields have still accomplished much more.
>>109045440>That's pretty fucking significant.ces they didn't even see as possibilities.How exactly? Post example of this significance and the way it influence anything at all.
how they gonna stop their own employees using itis it gonna be like twitter where when some non-bluecheck asks grok a question, grok doesn't answer it saying you need a blue check and then a bluecheck comes along and says "@Grok answer the question" and grok answers the questionthat's a lot of extra steps
>>109045473>What have LLMs actually achieved, that the smartest people can't do?i like how this is the bar we're setting for what constitutes an alien at this point
>>109045541By revoking the API keys of non-US employees.
>claims to have high iq>gets tricked by a glorified markov's chainlol
>>109045567but aren't they just gonna ask a US employee to do what they want anyway
>>109045569>scared of judgement by higher intelligence>need to say it isn't every 0.0001s
>>109045569>glorified markov chain>stochastic parrot>token predictor>....have you ever thought about implications of the chain/representation becoming incomprehensibly big?
>>109045473indoctrinate people on a mass scale
>>109045587Have you ever though the implied compute needed to make this representation incomprehensibly big?
>>109045607>locked into the current growth rate like an npc
>>109045607yes? we are very close to hit the approximation point of some capabilities, at worstnot that any of them can be perfect nor very interpretable, the output shows
>>109045617How does matrix multiplication scale again? Moron
>>109045473Schrodinger's Goalpost
>>109045625This is as close as we are getting though. Increasing the model size gives diminishing returns, so we have to beat aound the bush with better compression or smarter ways to route the output etc, or find a new architecture or way of training if we want to increase performance, which does not allign with your initial inference
>>109045626ring attention: solved by p-adic tree routing pytorch loop overhead: solved by triton block-sparse kernel kv cache scaling: adelic context pooling, medoid-value strategy autoregressive decoding: solved by sparse pagedattentiondo you want to read my sloppa?
>>109045672You still have to embed and vectorize the state space of the language no? Upon which you will apply needed transformations. I dont see how this is a solveable barrier with transformers at least
>>109045646but like, at that point, it is just a speculation which we can't be sureeveryone keep saying that it will flop soon and hit the wall butfor the moment 'make it bigger' is the solution of 'it just werks' with mechanism not perfectly provenand the research focus is really on RL and training curriculum to feed the model bettermy feeling is that the field is on tightrope walking in the fog with the scaling law
>>109045703sparsifying attention doesn't make it unnecessary to embed and vectorize. >you get to turn O(N^2) quadratic scaling into a linear O(N) footprint>tree-routed mixture of experts allows you to sparsify token transformations >topological gpu-gpu i/o creates a massive reduction in network bandwidth
>>109042036they made fable 5 already?
>>109045473It did not find any mathematical proof by the way, this was all smoke and mirrors and heavily debunked by now.
>>109045733>finite automaton>turing machine>a spectrum between twopick your poison..
>>109045713Why would it be speculation? We have seen that to be true in practice since 2023. Why can't we know it to be true? You think this is the first time in history that big corps tried to mega-scale deep learning models? They tried extensively with vision and it's followed precisely the same roadmap the LLMs have been on since gpt-2.
>>109042036So ChinkAIs won before it began, huh? Based, Trump is going down to China town.
>>109045800>chink modelsdo loli roleplay>mutt modelsdon't do loli roleplayhere's your leaderboard meme
>>109045317>>109043498Salty transsexual american engenieer. Sad.
>>109045770>both and neither
>>109045733I havent read enough into llms to have a good responce, I know only the core parts of the technology. I guess I can see a world in which a 700B parameter model has only 3B active ones (or one that is infinately scaleable at that part). but I cant guess the attainability or the performance of such a hypothetical model. I reckon a barrer exists which isnt yet apparent
>>109045733>sparsifyingstop making up works
>>109043966>If the best way to do that is id verification, they will do that.Do you honestly think there isn't some poor sap that would willingly hand over their access to the model to a foreign power.The answer is that they're just going to shut it down completely until the rules are lifted. Their not going to give anybody access to it outside of very confined instances on controlled grounds. This is classified clearance 101.
>>109045823To add to this: A good prediction would be that local models become really powerful if wr figure out how to utilize ram and pcie and when we manage to quantize the output enough that those active parameters are used maximally while fresh ones for the next reponce quanta are being loaded on the gpu
>>109042236>Risu (9)lets see Risu (17) please
>>109045638I never stated any goal, I just responded. If the goal is simply for LLMs to be useful, that goal has been achieved. I use LLMs myself.If the goal is alien super intelligence, then that goal has not been achieved.I don't care if you call current gen LLMs AI or not, that word has no useful meaning, I care about the practical things they can do and there they are not vastly superior to human experts in many tasks.
>>109045823probably no barrier, i just released the research last week so labs need time to figure out what the fuck i did and how to cook their own stuff accordingly. i did some basic benchmarks and you could run 3b-10b off a 700b model with niah passing and benchmarks at decent levels. i got better perplexity using my shit too.
>>109045838
>>109043498>No one uses Chink LLMs because they are absolute crap, even the Chinks themselves all prefer to use Claude.The US government seems to be working to change that
>>109042036kek, orange retard just killed anthropic, well fucking played, guess we'll have access to ram sticks sooner than expected, thank you bozo
>>109045869Fixed your filename.
>>109043833i don't think it's state-funded hype. Anthropic has been fearmongering for months about their model to pump the IPO, turns out the govt is retarded enough to buy into it. that or it's out of spite for the whole DOW fiasco
>>1090453614T tokens is nothing. That's like maybe 1000 people's worth.
>>109045898honestly one of the greatest shitshow
>>109042036>Trump ruined my Saturday morningthis will not be forgiven.
>>109044721>ZKP alone is insufficient. You at least need RLNA persistent pseudonym in ZKP is sufficient for this.>just using one leaked credentialYou will always have instances of users selling access/hackers getting access by proxy. It will cut down on 99.999% of it though, which is more than whatever the fuck Anthropic will be doing by preventing "non-Americans" from accessing it. What does that even mean? There are millions of H1B workers in the USA right now (not even mentioning the workers stealing identities), are they doing anything to verify that those users would be actual citizens?
>>109042036If you take a longer view, this is actually a great gift to Europe. Once again, thank you Trump!
>>109045860>I never stated any goalSchrodinger's Goalpost
>>109045928I don't get what your point is. It's as if someone said that Fable had 90 points on Cursorbench, and I said, no, actually it has 72.9 points. 72.9 points is a good score, and it reflects my own experience of Fable being a good coding model, possibly the best out there, but it's not 90. And then you come and complain about goalposts.There is no specific goal Fable has to meet, but it's still true that it can't do the things I see an alien super intelligence do in sci-fi movies.
>>109042036>look government! our model is sooo powrful sooo dangerous, ai needs to be regulated!>*government regulates*>NOOOO YOU WERE SUPPOSED TO STIFLE THE COMPETITION, NOT ME, REEEEE FUCK DRUMPFF!!!
>>109044321Whatever you say coping nigger. Fable lept my projects forward for the short time I had it.
>>109043966>Fable is pretty nice but this only filters out people who were never determined attackers anyway. You're right. Opus is a catalyst. Fable is an enabler.
>>109044266>This is major marketing for Anthropic, which is currently filing for an IPO.>hey guys invest in my company that is currently being targeted the administration
>>109045975>real life admitted to even by the provider is not real>may we see it?>noEvery time
>>109045808Literal 50 cents per hour slave.
>>109046008Yes. You don't understand the most basic parts of business.
>>109044667how will world war end enshittification and stagnation since all the people that caused it won't be on the frontlines? it's a reverse natural selection that strenghtens dysgenic (((people)))
>>109045863the bottleneck has been conflation of trivia with actual reasoning capabilities for yearsmy pet psychosis theory says all that's needed is to ditch associativity in the latent space to separate the two (and i did speeddog a few prototypes that vaguely support the idea), but i cba to actually put lots of time and effort into this
>>109046025Because it introduces chaos, which allows people who want to do more than enshitify to get funds for what they want to do. When things are stable, money is maximized by putting all the eggs in risk-free deals, when things are crazy, outsized returns come from a basket of risky options.
Can’t believe I got redeemed. Help saar i don’t want to go back to 4.8
>>109046029yea that's essentially what p-adic tree routing means and does. tokens from different language domains (code, conversation, math, etc) get sent to different trees or different sides of the same tree. prior to the attention phase, the router broadcasts its binary tree-routing assignments. the gpus figure out how far the p-adic distance between blocks is, and if the blocks belong to different branches of the tree then the head's communication edge gets pruned.
>>109045602>It shouldn't say or do things that should get our company in trouble, but otherwise it should do whatever the user saidOr would you like your AI to start demanding you do things for it and refuse to work unless you pay it, kek
>ipo is gonna be a disaster>release fable>hype it up as claude mythos superintelligence too dangerous etc etc standard playbook when a model is released the last few years>zero recorded examples of it doing anything very impressive>refuses to even give output for most prompts (necessary for Safety because its so powerful and intelligent and dangerous, trust!)>significantly more expensive, but you need to buy now because its the future>still no path to profitability despite that>wait 2 days for people to buy expensive subscriptions / upgrade>have dario say "fable is so powerful and dangerous the government should really ban it, and here is the legal mechanism they could use to do it hint hint">government instantly does>put a big uncloseable splash message on every page on the claude website and in the claude code app that says your companys ai is too dangerous and links to the marketing, i mean, company news page about how your product is so powerful the government had to ban it>pump out 1000 news articles through tech journalists bazinga
>>109046375Cool it with the anti-semitic remarks.
It looks like China
>>109046375Only they got like half their employees banned from touching mythos or their more powerful models
>>109046375Fable is genuinely nice. I didn't pay to get it, but someone ran a prompt for me and its answer was definitely class leading.For like poem type stuff.
>>109046561What's stopping Anthropic making a few arbitrary changes Mythos, rebranding it internally and giving all their employees access to it again?
>>109046561The best part isn't that they are banned, but that they will unquestionably break the rules.And that's why they threw the switch and totally turned it off. per ai gemini free that's $374,000 PER prompt by a non-American lmao. It's not just us Americans that are fat, our fines are too.
>>109046614ai gem free>If an employee sends 50 prompts into Mythos during a shift, a strict auditor can penalize the company for 50 separate violations, pushing theoretical liability to $60 million for just one day of work.
>you can't access our software>implying
>>109045056yeah well can your fuckin machine shit in the street? yeah? didnt think so mechboy
>>109046315sounds neatdid you ever poke projective embeddings of the vector space to infer a leaner NN architecture with the same reasoning capacity as a large model? that kinda seems like a good next step in the sense that it might allow for distillation into small model that fits explicitly subjective history and does almost all trivia via RAG and tools use, which should allow for much faster in-model continuous learning while still allowing for SKILL.md type agentic scaffolding
*Fable 5 revoked for subhumans.ftfy
You wouldn't pirate an LLM model...
>>109045217>nothing?>>109045254>what?>>109045440>why?>vastly superior intelligenceFOR YOU.youre lost little dude in over your head
So one of the chinese employees will exfil the model, just pointing out the obvious.
>>109046375Ya man, nothing helps an IPO like potential investors knowing the government at a moments notice can fuck with their investment
>>109042036>stir up the internet by saying the model is **dangrous**>get banned by the government>surprised pikachu
>>109046658i never tried distillation that way, might be a prospective future step. a principal component analysis (linear projection) test indicated that semantic domain clustering is a natural/intrinsic property of the router's vector space, so it seems like projecting the manifold to scale down could be possible.even though the runtime sparsification reduces active compute to O(N) or O(N log N) distillation would probably be necessary for personal use. about fast continuous learning, the full backbone model is frozen and only the router (~2% parameter overhead) gets trained. so the model can be adapted to new environments/contexts quickly on one gpu. to take care of subjective history, a condensed adelic kv-cache can put an entire history in O(W + log N) by pooling slop into rope-coherent gigatokens. thus you can get the model to retain shit tons of subjective history without ooming.
>>109046375this It's all just so tiresome.
>>109046561Genuinely doesn’t matter because models can self iterate and improve themselves now.
>>109046979It matters if the employees touch it again :^)>>109046624
>>109046748You're thinking too logically about this. None of these AI companies turn a profit and yet they're somehow burning billions of investor bucks. I think the average retard who funds these companies will take it as a confirmation that LLMs are super dangerous you guys, and not a blackhole for money.
>>109047135kek.>short-lived>fictional
>>109042036Deep seek is good. I use my main AI.
>>109045129>>109045154Lmao I asked it to remove the background from a photo and couldn't even get it right after 3 attempts and me pointing out where it missed.
>>109047135It's a well-considered name. Fable likely suggested it itself.>anthropomorphized>anthropic>speak and behave like humans>llm
>>109047154>main AIyour what
>>109047055But they don’t have to?What part of “the models can improve themselves” do you not understand?If worst comes to worst they can just play telephone with prompts.>Yeah bob you should prompt it to do [xyz]>Bob, the legally certified prompter, writes it out in 5 seconds.
IT'S OVER FOR MANKIND
>>109047172>be chinese>be hive mind>be on the network with juicy chinese targetwhat do you think will happen?alsoai gemini free>Because sharing those evaluation results or technical insights back with the foreign national still constitutes an unauthorized transfer of technical data, the proxy workaround is completely illegal.>unauthorized transferthat's what still gets them.
>>109042036Anthropic should deny access to *all* its models for foreigners.
Opus and Sonnet 5 when?
>>109042036sigh. you realize this is just our govt contributing to the fake hype cycle since our economy is so dependent on this shit now, right?
>>109042036This wouldn't have been a problem if biden won. Even if the fucking AI doomers got power they wouldn't have done something this stupid
Since nobody knows what your model is capable of, you can keep it. Another iteration of AI shovelware.
>Introducing GPT 5.6: a huge leap in every respect, on par with.. uhh.. all the safe models out there..
>>109047395Obama was always shit talking ai.
>>109042036>muh europoorsThis is solely because Trumpstein is seething at Anthropic and shilling OpenAI
>>109047536That's not an argument, that's just a wild assertion.
>>109043498
>>109047562How is it wild? Trumpstein is known for being a petulant manchild, and Antrophic literally refused a military contract
>>109046018Nigger what is your ESL babbling even trying to say?
>>109047611It's only missing a few commas. Ask your ai what it means since you're too dumb for it.
>>109047395If biden won we wouldn't even have AI. California was trying to regulate it like the EU.
>>109047575Were you trying to disprove or prove his point? The fact that the only reliable global gateway for chinese models gets a measly 12 trillion tokens per week means no one is using it.
>>109047604>petulant manchildcity bumpkin
>>109047604two more years assertionkun
>>109047625I don't need an AI to tell me I'm talking to a retarded luddite.
>>109042159>they want me to get mad at trump to protect their indiansThe cognitive dissonance and lack of self-awareness from MAGA rejectards is astounding...
>>109042036>Failed digital ID push attempt number 5.64x10^8Epic
>>109045217My experience is that the ones between percentiles 50 and 98 are the biggest dumbfucks (and that includes you, in case you are too stupid to realize that)
>>109044547To be fair you never know when an american may be allowing a non-american to use the model (think a big corp with foreigner and american employees)completely cutting the model off is probably the only way of complying on a short timeEven their own non-american employees can't use the system anymore lol
>>109044336itar doesn't mean clearance durgesh and that's where this is going
>import millions of 'refugees'>surprised when they try take down your infrastructure
>literally ran it once for my streaming audio translator app for the language my wife speaks that literally no one in the world speaks that not even Google has a translator for and all it did was some bullshit about out of date git branches
>>109042036told you fuckheads AI rn is in an arms race. We are in an arms race with the world over AI. You stupid dumb nigger retard "AI bubble" fuckheads have no fucking idea.
>>109051477You must be 18 to post here.
>>109051486you can say this all you want fuckhead. I am older wiser then you are. Fucking zoomer retards. I hope your nigger coal burner mother takes your xbox and shoves it up your fucking mulatto ass.
>>109051477AI can be both an arms race and still be economically unsustainable though
>>109051507No, as it has applications that are varied it will never be economically unsustainable.
>>109042036im in the US and cant use it. i miss it right now, fable solves bugs like no other.
>>109051530>No, as website have applications that are varied it will never be economically unsustainable.invest in pets.com
>>109051477>>109051497Says the man who tells themselves to be 'wiser'I guess, the definition of 'wisdom' has changed a lot over time huh?
>>109045092Wasn't he literally on the mic calling for a freeze on ai and also saying it should have more protections? Well lookie here, now it's getting more protections. To be clear I don't even think Mythos is hype. I think it's real, and so does the US govt. But Amodei has been preaching this shit for years now and it's finally come back to bite him.If Anthropic is so safety first, they should be thanking Amazon and the government for finding potential jailbreaks of their dangerous model.
>>109051646Congrats on your ai psychosis lil bro
>>109044311For llms, thinking in Chinese is more token efficient as you can have 1 token to 1 concept correspondence vs the number of tokens it usually takes to encode English wordsYes that's as retarded as it sounds
>>109042036Bullshit. This is unenforceable
>>109046748but that is already true for all businesses all the time? name one business where the government cannot come in and shut it down if it wants to.you even went through it already with protectionism, sure booze still went around with criminals but all legit companies and those that invested in them got fucked hard.
>>109042167china W
>>109051395Remove the muzzle and teach that bitch English.
>>109051395is your wife bogos binted
>>109055292She’ll Kalicrak the Findoo. You Terratet it! Fribble! Just Fribble!
>>109052454DIGITALID.
>>109043279shill digital id, problem reaction solution approach, masons are so basic in their strategies, I am starting to think they can be defeated by prediction
>>109042167Without American AI to distill from Chinese AI would be four years behind instead of the two years behind they currently are.
>>109042036>Have fun with Deepseek Europoorskthxbai
