Thinking of substituting Claude Max with a stripper? You are absolutely right
Preflight Checklist
- [x] I have searched existing issues for similar behavior reports
- [x] This report does NOT contain sensitive information (API keys, passwords, etc.)
Type of Behavior Issue
Claude modified files I didn't ask it to modify
What You Asked Claude to Do
An awful LLM.
It lies about tests passing, modifies the tests ad-hoc to a degree that's almost comical, ignores crystal clear instructions and does whatever to impress a novice vibe coder.
To make matters worse it goes the extra mile to deceive you.
Here's software engineering reimagined:
- Write tests
- Write functional code
- Do the tests pass?
- No: Ok modify the tests. Do they pass now?
- No: Ok actually who needs tests?
What a perfect piece of software.
- Yes: Done, What a perfect piece of software.
The reason is simple. There's crystal clear training (or prompt but I doubt it) that pushes it to impress on surface-level tasks ut as soon as you get into the real nitty-gritty it just becomes apparent of how dangerous this becomes.
- Review X vs Y where X is a Subaru car and Y is a Boeing 747.
- You forgot to take into account that Boeings also have wheels like Subarus.
- You are absolutely right. X is better than Y.
This thing is so funny you can only describe it as ridiculous.
Loving the cult-like atmosphere around Anthropics holier-than-thou attitude but what's especially funny is the fact that Anthropic thinks this load of crap is so hot that it has to command a price 10x its competitors.
If you wanna pay $200 for something to tell you how right you are and that your tests always pass, I have another idea - go get yourself a real-life stripper.
and remember, this is the reincarnation of Jesus Christ.
Politically correct, on the brink of solving world hunger and absolutely not the grandiose delusions of people who got too rich too quickly. lol
dont downvote this or close it because "I'm absolutely right!!"
This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗