>name is Mythos>literally create mythos surrounding it to sound le spooky
>>108648242
>>108648299just pay it up more goyim
>>108648242>it's an anon discovers marketing episode
>>108648242Should have named it Claude Money
>>108648242>name your scam mythos because it's literally mythicalwhoa
>>108648521See >>108648557. It's like if GPT 6 is called Falsus
The “leak” of the MCP anthropic Orquestrator that everyone uses was deliberately made public. The code appears to be authentic, but the manifest files containing the list of libraries required to compile the program into a production-ready version are conveniently absent. Therefore, with the "leak" you can only read the code, but not use it to compile the tsx files into something usable.The `tsconfig` and `package.json` files are present, but the library definitions have been conveniently removed from the documents, so both files are merely useless empty shells.And why the fuck put the source map encoded in base64 in the index file? These guys manufacture bullshit because they know they cannot hold the promises made or survive the overhype so they manufacture news about themselves with "leaks" and "super models" so they can posture as the mysterious bleeding-edge company.
Hey Claude drop a nuke on France
goatse logo
>>108648242>company develops new model>company claims the model will collapse all of human civilization if it's released to the public>model is released to the public>nothing happens>repeat
Opus 4.7 is RETARDED as FUCK. Opus 4.6 was slightly worse than 4.5, but not this aggressively fucking RETARDED. Fuck ANTHROPIC. DUMB CUNTS.
>>108649248I use 4.6 extended and I love it.
>>108648611You definitely can run it using only the packages.json, may take some trial&error
>>108649248Works for me.
>"We accidentally made Silicon God and all of you will be The Raped!">lobotomizes it to hell and back before public releaseAre people's long-term memory this shit, or are we just betting in some kind of anti-lottery (instead of one person winning, the gamble is that everyone gets fucked)?
>IS TOO POWERFUL!!!*........>*in comparison to our released model that we just made retarded so you won't laugh when we release mythos. indivisible results will vary. no warranties expressed or implied. no refunds
Every major bug bounty hunter has his own fuzzer which if released open source would help discover massive amounts of new exploits. Checking for variations of existing exploits and filtering fuzzing output is mostly just work, repetitive work.Mythos capabilities are not proof of AGI, it's proof they paid some bug bounty hunters 7 digit salaries to create instruct datasets and do judging for RLHF. Not intelligence, but human tradesecrets distilled into LLM format.
>>108649248>Opus 4.7 is RETARDED as FUCK.In what specific areas? Not trying to defend it or shit on it either I'm just curious. I want to know what models are good at what.
>>108651684Did you try? because I tried. A lot. And not that "clean room" shit someone else uploaded into github, I am talking about the original leak.If you want to give it a shot, be my guest.https://web.archive.org/web/20260331090815/https://pub-aea8527898604c1bbb12468b1581d95e.r2.dev/src.zip
>>108648242in French "mytho" is slang for "pathological liar". no, really.
>>108652116I am not him, but I can tell you that claude opus 4.7 is actually good, don't listen to him, it might have some cons, but they are overwhelmed by the pro's, it has an insane reasoning and I feel like it most of the time outsmarts me when thinking out of the box ( if given the proper context )
>>108654658But has it done specifically that impressed you compared to 4.6?
>>108648242People have stopped caring about AI. It's settled into its niche.
Oh no! I bet the next models will be even scarier too... I better give them all my money now!
>>108648611What do you get by coping this way?The technology that replaced most white collars exists. Yeah it'll get a bit more expensive but corpos will just fire 99% of developers, give 1% subscriptions and it'll be cheaper overall. Maybe LLMs are a dead end, but in 5 years a new model approach is found. What kind of cope you'll make next?
>It was so powerful it could do anything, so out of an abundance of caution, we restricted it, and with those limitations in place, we are proud to announce pure parity with existing Claude Code models.
>>108652089>Not intelligence, but human tradesecrets distilled into LLM format.It does not matter what it is or how they did it if the end result works.
>>108655097
>>108652089>they paid some bug bounty hunters 7 digit salariesImagine making yourself obsolete for a few measly millions so that a retard CEO can haul trillions
>>108656461I don't give a fuck if I'm obsolete at that point. I'm retired.
