Claude Mythos is out
>release a new model>"so powerful, world and economy end soon">is expensive slot machine for code>waste 6 gazillions monies>throttle and lobotomize it>it's now worse than previous models>next model will seem better>slash usage to a quarter>gaslight users, "skill issue">profit? (negative 5 gazillions monies)>repeatyou niggers will live this forever
i want to dive into vibe coding again. what is the current zeitgeist? how is the harness, can i still roll with mcp and ignore skills?
>>108471800Mcps and skills target different things imo. Skills are just mini CLAUDE.mds that get loaded when needed. Just a short natural language description of how to do something in your workflow, without needing to copy paste it into a prompt.
>>108471784>>waste 6 gazillions moniesI wish that was all it did. Codex put memory corruption bugs in my C++ code!!! It wrote this sprawling mess of superfluous singleton classes. I feel like it violated my program.
is this why CC has been so retarted? was it worth it?
>>108471999Its not just me? Ive found opus doing some stupid shit recently like I tell it some fact about the project and its asking me two prompts later even though I told it. Its been a week or two its felt off.
>>108471681imma wait for distills then
>>108471784Cope Luddite. You will be replaced
>>108472751>Mom, you don't understand! Everyone will be unemployed as well. Just two more weeks!Sorry buddy, couldn't be me.
>>108472751
>>108472824That's how I always looked at my code when no one on Stackowerflow answered.
>>108472838looks like programming is not for you then.
>>108471784They're doing the fake leaks meme now in addition to shitcoin pump and dump marketing. >ANTHROPIC DOESNT WANT YOU TO KNOW HOW POWERFUL ITS NEW MODEL IS>INSIDER DOCS LEAK ANTHROPIC IS TERRIFIED OF THE POWER OF THEIR NEW MODELi will take scam Altman 1000x over this fake and gay hype marketing
>>108472882nah i think im just going to keep using ai :^)
>>108471784vibecoding will always be shit because if they make it too good, then they won't make much moneyit always has to give you meh results so you keep wasting tokens, kind of like google search
>>108471971Did you tell it singletons are illegal and you'll go to jail?
>>108472927here got a book for you
>>108472927based. keep on btfoing these luddites.
>>108471784ESL
>>108471999>>108472531Claude Code has been lobotomized recently. It's fucking retarded now, I don't know what happened. Replies are faster than usual and they're short, shallow, and dumb. It's like they bimbofied it without the sex appeal
>>108472942
>>108472824>Car drivers looking at their cars's innards after they run out of money to pay the mechanicWorks for lots of things bozo
>>108472977Maybe too many new users so they gotta cripple it to conserve resources?
>>108473003Someone said they were training a new model and a few days later they announce Mythos. I think it may have something to do with it.Others were saying it was because of Iran and the government was using all compute power but the government has their own servers, so unless they started redirecting consumer GPUs over to the government servers I can't see this being a thing.Maybe it was an influx of new users, who knows. All we know is they're not being transparent about it.
>it's retarded wow it's almost like "Artificial Intelligence" is a misnomer and there is no intelligence or reasoning. REMINDER>Intelligence: The ability to come up with novel and USEFUL ideas from pre-existing ones. EXAMPLE:>give AI all math knowledge prior to Isaac Newton>AI should be able to WITHOUT HUMAN INTERVENTION >Identify the gaps>Come up with relevant questions>Solve those questions>Create calculusAI has not been shown to do anything REMOTELY like the above. There is no intelligence. It can only generate a hazy cloud of probable information that a human needs to sort through.
>>108473048@Grok make him shut up
When will I be able to use it?
>>108473063I got you fellow human>>108473048>wow it's almost like you just described how EVERY human "invents" anything>REMINDER>Intelligence: The ability to come up with novel and USEFUL ideas from pre-existing ones>EXAMPLE:>give a human baby all math knowledge prior to Newton>raise it in a cave with zero books, teachers, or tools>it should WITHOUT ANY HUMAN INTERVENTION invent calculus>no human in history has ever been shown to do anything REMOTELY like that either>there is no intelligence. just a hazy cloud of probable information that needs society/books/teachers to sort through>congrats retard you just proved humans aren't intelligent either>or maybe your definition is dogshit and "intelligence" is pattern-matching + iteration on existing knowledge>which LLMs already do better than 99% of /g/
>>108472838If that was true beyond the first year of your career, you need to reset your baseline. You managed to slip into a job you're not ready for. You can learn, and fake it until you catch up, but as someone who's been there, you're probably better off looking for similar paying jobs that deal in areas you have actual expertise.
>>108473123WOAH - why are you comparing AI to human babies? I thought AI was the superintelligent GOD who can replace ALL human work? why infantilize it? >iteration on existing knowledge>which LLMs already do???no it doesn't. LLMs have never came up with anything novel and useful that wasn't already in it's dataset.
>>108473227>>[his post number]>WOAH bro you just completely dodged the point lmao>you set the bar at "invent calculus with ZERO prior knowledge/society/books/teachers">then cried when I showed a human baby under the exact same conditions also can't do it>congrats you just proved humans aren't intelligent either by your own standard>no one here called AI a "GOD" or "replaces ALL work" you schizo>we said it's pattern-matching + iteration on existing knowledge>which is literally how EVERY intelligence works, human or not>LLMs don't pull shit out of a vacuum... neither did Newton, Einstein, or your mom when she invented you>examples of novel useful shit LLMs produced that wasn't copy-pasted from dataset:>- FunSearch (DeepMind LLM) discovered larger cap sets than any human in 20+ years + brand new bin-packing heuristics that beat every baseline>- AlphaTensor found faster matrix multiplication algorithms, improving on a 50-year unsolved problem>- multiple LLM systems now generating original math conjectures that get proven or falsified>all post-training, all verified novel by actual mathematicians>cope: "b-b-but it was in the dataset somewhere!!!" is literally the same excuse flat-earthers use>keep moving the goalposts tho, it's entertainingThat is groks reason to you
>>108472996my car can get me from point A to B without hallucinating
>>108473306You know I think I should start using grok more for reply’s. No point in thinking for myself. Just have grok do it for me.
>>108473309nope, you gotta steer it else it hallucinates into the sidewalk
>>108472785It sounds like a cope bro. Coders aren’t required anymore. AI won.
I guess it’s my fault for expecting this thread to actually contain discussion of the model.Is anyone using it? How is it?
>>108473736You're replaced, why do you care how it is ?
>>108472824Such is the life of a clanker whisperer.
>>108473736Yeah bro let's leave this thread in the catalog just in case anyone gets access to it in the few months until it's released to the general public but only bump it when it reaches page 10!!!
>>108471681fuck everything about ai
>>108471784Can't wait to see this copypasted in every ai thread until time ends
>>108472936good pointbut its bretty gudd now THOUGHEVER
>>108471681Not for me. Are you on the $200 plan? I wonder if it's because I'm only on the $100 one, or if it's a gradual rollout
>>108471784fpbp
>>108471800I have been using Claude at work for a month or so. We had an "expert" who has built his entire startup around doing everything using AI. His tips were, as far as I can remember:>Spend some time writing good instructions in CLAUDE.md so you don't have to correct the AI as much.>Use the planning mode to plan out bigger changes.>Have another AI, he used Codex, review the plan devised by claude.>create sub agents to keep your context clean>remember run a /clear between every unrelated task to keep Claude from getting confused>don't be afraid of running several AI sessions in parallel so you don't waste time waiting for the AI to finish whatever it is doingBefore trying Claude, I was very sceptical about the AI hype, but it really impressed me. It's ability to reason about code and systems has on occasion been better than my own. It's major problem is its lack of common sense on an architectural level. If you let it actually design things, it often makes a lot of retarded decisions.
