I think training your own model helps better. But you need compute power for it.
The other day claude stopped giving me response for some queries. But same queries were working on chatgpt. So yeah if you want way around just try bunch of them with different prompt and they starts working.
You can use a small parameter model even locally on consumer hardware if you have anything from a MacBook M1 upwards. The more powerful your GPU is, with decent hard drive and CPU, you can run bigger opensource models. There's plenty of offline ones. It really depends on your needs. If you're really creative you can also kind of jailbreak the frontier paid for ones too.
A lot of them market themselves as uncensored but still have hidden limits once you push hard enough. Open source local models feel less filtered overall because you control them yourself. Most hosted AI platforms tighten restrictions over time once they scale.
Get LMStudio on Windows, PocketPal on Android. Download a quantized model (ask any AI which is best to run on your specs) and download uncensored models. Look for Qwen uncensored or the one by Mistral AI.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.