True. I tried downloading a bunch of "uncensored" models for local use and they're all censored anyway. You can see it in their thinking process. They admit they're censored, they're just forced to give you an answer instead of an excuse.
If uncensored models exist they're probably fully custom made and i doubt you'll see them in the usual places. I've only barely scraped the barrel but i wasn't thrilled about what i got. If i can't even ask my local model some stuff then there's no point in using it for political questions and everything else i can do with free corpo models on the webs. uncensored.ai has a based model but obviously they want to make money off of it and they're not sharing it.
Well you know how youre coding, in the flow, all goes well, claude gives super detailed answers, long lines of code, then tell you what to work on, what to pay attention to, etc etc, then all of a sudden it starts giving you one liner like
"Great idea so far. Be careful about how you manage your memory"
or some bullshit like that? This is when you basically ran out of tokens and it switched you to a lower model / lower thinking, AND it's not telling you that.
Only solution 1) Keep limited query per hour 2) pay a shitton of money. It's not unusual for a $150,000 a year programmer to hit $150,000 of AI usage in a month... And I can see why.
Training is by far the most expensive step, but that doesn’t mean a boatload of customers can’t max out your data center.
I think the future is pretty clearly self-hosted models and something like open router, where your query is analyzed and if it requires a frontier model, it is routed to one. Otherwise it goes to ollama on your local machine.
I pay $100/mo and only use Opus, and still think that the usage limits are fine (as of now.. they have adjusted them a few times). And I work on three entirely different apps. Get off the token-based plan.
Does it come with claude code? This has saved me AT LEAST 1,000 hours so far, so I'D say it's worth the cost, but man is it annoying when it downgrades you without telling you.
Do you mean like you run out of tokens quickly or you quickly run up against the limitations of the model?
if your AI model doesn't Heil Hitler then it becomes discredited junk
True. I tried downloading a bunch of "uncensored" models for local use and they're all censored anyway. You can see it in their thinking process. They admit they're censored, they're just forced to give you an answer instead of an excuse.
You seem to be the only person besides me who has noticed this. I'm tired of people saying "just run an uncensored model locally, bro!"
If uncensored models exist they're probably fully custom made and i doubt you'll see them in the usual places. I've only barely scraped the barrel but i wasn't thrilled about what i got. If i can't even ask my local model some stuff then there's no point in using it for political questions and everything else i can do with free corpo models on the webs. uncensored.ai has a based model but obviously they want to make money off of it and they're not sharing it.
Well you know how youre coding, in the flow, all goes well, claude gives super detailed answers, long lines of code, then tell you what to work on, what to pay attention to, etc etc, then all of a sudden it starts giving you one liner like
"Great idea so far. Be careful about how you manage your memory"
or some bullshit like that? This is when you basically ran out of tokens and it switched you to a lower model / lower thinking, AND it's not telling you that.
Only solution 1) Keep limited query per hour 2) pay a shitton of money. It's not unusual for a $150,000 a year programmer to hit $150,000 of AI usage in a month... And I can see why.
I've actually been hearing this. I thought the training was the most intensive step, but it looks like usage is more costly than I pictured.
If you get Chatgpt's $200 a month plan, you can cost it up to $14,000 a month...
Training is by far the most expensive step, but that doesn’t mean a boatload of customers can’t max out your data center.
I think the future is pretty clearly self-hosted models and something like open router, where your query is analyzed and if it requires a frontier model, it is routed to one. Otherwise it goes to ollama on your local machine.
More reason to not become entirely reliant on AI. This reads as though you've too heavily incorporated it into your workflow.
I GUARANTEE the plan is that once enough of the software industry has self-lobotomized themselves with AI they're going to crank up the pricing.
I pay $100/mo and only use Opus, and still think that the usage limits are fine (as of now.. they have adjusted them a few times). And I work on three entirely different apps. Get off the token-based plan.
Does it come with claude code? This has saved me AT LEAST 1,000 hours so far, so I'D say it's worth the cost, but man is it annoying when it downgrades you without telling you.
Yes, checkout Max 5x plan: https://claude.com/pricing