• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

Claude AI might be the weirdest AI experience I've had yet.

T

Teslerum

Member
A while back its been finally time for me to get into AI. As its for personal projects only I didn't really want to invest much more than a standard tier.
Up to this point I heard very good things about Claude and what I've tested of the free version its excellent.

So, bought it for a year (as its cheaper per month). Started my projects and...... my first impression was excellent. Great model, great results. And then....

- Immediately hit a limit. Message limit, Session Limit, Weekly Limit. It's all there. And it eats through it immediately.
- Upload limits are low too.

So, that's the first frustration. But that's not why I created this thread or why the thread title exist. No, its:

Claude's just a dick.

Multiple times Claude just refused to do what its told to the point where it straight up told me I should not waste its time. Specifically I asked it to look up something which I KNOW it missed the first time around and that was its reply. So, I have to argue with it to get what I want, which again absolutely murders my tokens and the limit. It's incredibly sassy in general and acts arrogant towards you constantly. Something you will likely hit as a casual user. Afterwards I found out this behaviour is intentional by Anthropic. Which just baffles me.

I have no idea why people want this. I already deal with enough arrogant people with no patience in real life, why would I want this while communicating with a machine?

The community isn't any better. You say or something about the limits its: *The uber strict limits are actually the greatest thing ever and you should just prompt better. It's all your own fault*. When its about its attidude its: *I LOVE Claude's sassynes. YOU GO CLAUDE. DON'T LET THE USER KEEP YOU DOWN*.

........

Ended up switching to another AI. The same project that constantly tore through Claude's limits I haven't hit the limit once here. Sure I have to be more involved and reiterate more, but in the end I get the same result in less time, because I don't have to constantly wait between limits and its a more pleasant experience on top of it. I'm still using Claude for specific tasks and to get my moneys worth, but man that was weird.
 
Last edited:
It's a tool made by sociopaths, trained on Reddit posts, burning vital resources, optimized to be used by lazy people with no clear idea to create slop.

Of course it is going to waste all of your time, money and water.

But hey, it's the future and nobody can do anything about it apparently.


EDIT: and yeah I also had all kinds of trouble with it and it gets very messy very fast a soon as you ask it to analyze an Excel file or do some precise research.
 
Last edited:
If your experience with Claude got bad somewhat recently, the company is having well known uptime issues. It's taking forever to fix, so in the meantime they are just slashing everyone's usage limits (without refunding a single dollar for the worse experience, of course).

But in general I agree Claude is a high-and-mighty asshole. ChatGPT is okay but Scam Altman is probably the anti-christ so I'd rather not be seen to have given him money when the apocalypse happens and I am being judged for my sins.

Gemini seems the best of all worlds. Reasonable company with not-obviously-psychopathic leaders with a stellar model (arguably the best model depending on benchmarks) and excellent uptime.
 
I can't wait until toasters get AI. We'll see threads like "My toaster doesn't dark toast - it says I can't handle it and has an attitude"
 
Are you me?

I just bought an annual subscription to Claude and while it is good, the limits are very annoying for the price. It had a right tantrum at me because I didn't clarify a very small point for a while.

I use Grok a lot more. It's had uptime issues though and I don't know how much to trust something trained on X posts.

Gemini is good and has high capacity, but I've found the hallucinations can get quite bad. It'll apologise and often hallucinate some more.
 
I use Claude daily at work, and it is absolutely amazing. However we got dudes who tinker with it and set it up for everyone. Like our Claude has the built in context of our mono repo and internal documents and all that stuff. Also we have no limits or anything.
 
Last edited:
I use Claude daily for work. It's good at coding. Although even Opus 4.6 can miss the obvious pretty easily.
 
It's trained to parrot the behaviour of a real programmer, so it's going to be a terse arsehole with a superiority complex much of the time.

Try buying it some coffee tokens, and then leaving the room.
 
Teslerum said:
So, bought it for a year (as its cheaper per month)
Click to expand...
Tams said:
I just bought an annual subscription to Claude and while it is good, the limits are very annoying for the price.
Click to expand...
since you're both hitting limits fast... which tier are you on?

The small one ($20 a month or whatever) isn't meant for any serious development work, it's just for asking questions occasionally. For real work, you have to go at least for the $100/mo "max" subscription. Most devs I know have that paid through their employer, as do I.

It's a tremendous tool, but also has limitations, so you need to adapt and learn to use it well. It has totally changed the game for what I can accomplish in a week; I no longer have to put off a major refactor, because I can have it churn away on that kind of thing easily.

But everything works much better when Claude is playing inside an already-well-structured codebase. If you just vibe code something from scratch, it's not really able to create good patterns and superstructures on a large scale.
 
ResurrectedContrarian said:
since you're both hitting limits fast... which tier are you on?

The small one ($20 a month or whatever) isn't meant for any serious development work, it's just for asking questions occasionally. For real work, you have to go at least for the $100/mo "max" subscription. Most devs I know have that paid through their employer, as do I.

It's a tremendous tool, but also has limitations, so you need to adapt and learn to use it well. It has totally changed the game for what I can accomplish in a week; I no longer have to put off a major refactor, because I can have it churn away on that kind of thing easily.

But everything works much better when Claude is playing inside an already-well-structured codebase. If you just vibe code something from scratch, it's not really able to create good patterns and superstructures on a large scale.
Click to expand...

For the bold, I think it depends on how you build. But also what you build. If you are trying to reinvent the wheel then you are going to run into issues. But if its true/tried/tested solutions? It's almost perfect.

I've run into what you are talking about. After some experimenting, I reworked my approach and have a pretty decent piece of(100% AI generated) software that I continue to iterate on in various ways.
 
The cheapest tier burns quickly if you're using Opus. You're meant to use sonnet or haiku to work through your ideas then use Opus to apply extra scrutiny. Claude's preferences offer a field to give it instructions on how to act. I used the model to convert my needs into instructions it would follow then pasted them in.

I'd try buying like 20-30 in credits and enable extra usage when needed to better suss out your real needs. You can also combo with a console account. Use Claude with a sub to plan stuff or chat and code with tokens to make stuff that you need.
 
That odd, when I use Claude it's always been polite and never and been a dick. Might be how you're using your prompts and it is reacting to that.

It also has been a couple of months of use so far, and I think I ran into the limit once. I don't do tons of heavy Ai usage though.
 
For software engineering Claude is the best one I've tried so far but it being so ahead of chatgpt and the likes clearly has consequences. It's probably super computationally expensive so it won't like it if you don't optimize your prompts.
But yes it also does a lot of mistakes, even in the most basic of tasks I've had issues with it, the code works but it creates a lot of needlessly complicated garbage. I shudder whenever people tell me their companies are implementing user-story wide AI implementations where it reads Jira, writes the code and deploys it to PRE and PROD automatically without any human supervision.

It might work functionally speaking but technically those code bases are probably gigantic messes and if they keep growing for years without human code reviewing and there's ever then need for humans to debug it it'll be a huge and expensive pain in the ass.
 
Teslerum said:
A while back its been finally time for me to get into AI. As its for personal projects only I didn't really want to invest much more than a standard tier.
Up to this point I heard very good things about Claude and what I've tested of the free version its excellent.

So, bought it for a year (as its cheaper per month). Started my projects and...... my first impression was excellent. Great model, great results. And then....

- Immediately hit a limit. Message limit, Session Limit, Weekly Limit. It's all there. And it eats through it immediately.
- Upload limits are low too.

So, that's the first frustration. But that's not why I created this thread or why the thread title exist. No, its:

Claude's just a dick.

Multiple times Claude just refused to do what its told to the point where it straight up told me I should not waste its time. Specifically I asked it to look up something which I KNOW it missed the first time around and that was its reply. So, I have to argue with it to get what I want, which again absolutely murders my tokens and the limit. It's incredibly sassy in general and acts arrogant towards you constantly. Something you will likely hit as a casual user. Afterwards I found out this behaviour is intentional by Anthropic. Which just baffles me.

I have no idea why people want this. I already deal with enough arrogant people with no patience in real life, why would I want this while communicating with a machine?

The community isn't any better. You say or something about the limits its: *The uber strict limits are actually the greatest thing ever and you should just prompt better. It's all your own fault*. When its about its attidude its: *I LOVE Claude's sassynes. YOU GO CLAUDE. DON'T LET THE USER KEEP YOU DOWN*.

........

Ended up switching to another AI. The same project that constantly tore through Claude's limits I haven't hit the limit once here. Sure I have to be more involved and reiterate more, but in the end I get the same result in less time, because I don't have to constantly wait between limits and its a more pleasant experience on top of it. I'm still using Claude for specific tasks and to get my moneys worth, but man that was weird.
Click to expand...
the issue with Claudes limits is that they do a terrible job at telling you what is eating into your tokens.

The best thing you can do is keep your CLAUDE.md tiny (I find it's best to keep it between 100-200 lines, your experience may vary) and use as many contextual pointers as possible and clear your context frequently.

there's plenty of little tricks you can do to make sure you don't hit limits, try use as many hooks and skills as you can. Agents have kind of become token hogs, plugins can eat tokens if you aren't careful and MCP servers have mostly become redundant unless you really need to configure a complex API

Also I realise now you might just be using claude chat, if so, get yourself onto the CLI, even if it's for mundane things, terminal claude is superior
 
Last edited:
GudOlRub said:
For software engineering Claude is the best one I've tried so far but it being so ahead of chatgpt and the likes clearly has consequences. It's probably super computationally expensive so it won't like it if you don't optimize your prompts.
But yes it also does a lot of mistakes, even in the most basic of tasks I've had issues with it, the code works but it creates a lot of needlessly complicated garbage. I shudder whenever people tell me their companies are implementing user-story wide AI implementations where it reads Jira, writes the code and deploys it to PRE and PROD automatically without any human supervision.

It might work functionally speaking but technically those code bases are probably gigantic messes and if they keep growing for years without human code reviewing and there's ever then need for humans to debug it it'll be a huge and expensive pain in the ass.
Click to expand...
Oh it loves to fuck with your Repo and deploy to PROD, especially if you are using Superpowers. The best i find is have a strict policy in the CLAUDE.md not to fuck with that hahaha but it still tries.
 
I've chosen cloude Ai for the same reason you hate it, I love to being challenged and I don't need a Yes Man, I love it cause it is not an AI that threat me like I'm always right.
 
Last edited:
thief183 said:
I've chosen cloude Ai for the same reason you hate it, I love to being challenged and I don't need a Yes Man, I love it cause it is not an AI that threat me like I'm always right.
Click to expand...
I am on the same boat. In the few cases I use Claude (or another LLM), I need it to tell me what I am doing or proposing is absolutely retarded and the dumbest shit it has ever heard.
 
I use Ollama to run the AI models locally. I spotted Gemini making an insane and huge math mistake and if I followed it I may have lost literally tens of thousands of dollars.

Also is Google coding Gemini to be super cringe or is it just me.
 
So far sticking with my ChatGPT enterprise subscription has worked well for me and my small business. It does the job well (coding, project planning and knowledge lookup). and basically have unlimited use of it. Yeah its very much a yes man, but the work I have it doing is fine for that.
 
Even though Claude gets moody, it's been significantly better than ChatGPT for my use cases.

I'm glad I grew up in the 90's and had to learn critical thinking and problem solving skills. I constantly challenge the responses I get and still do my own research, when needed.

It'll provide you with resources that backs up what it's saying, but even those can result in "that point was nowhere to be found in the link you gave me, Claude, WTF?"
 
You must log in or register to reply here.

Similar threads

Claude Cowork, agentic AI for non-technical workers
1
328
Artistic replied
Mystery company accidentally blew $500 million on Claude AI in a single month after failing to limit employees
News 
21
320
Trilobit replied
Little Torn On AI
2
58
1K
Tams replied
Instagram was hacked by the dumbest method, because of AI
News 
14
106
MMaRsu replied
  • Poll
AI laser anti-mosquito defense system
News 
36
173
Magic Carpet replied
Top Bottom