Comment by anarticle
24 days ago
Download models you can find now and forever. The guardrails will only get worse, or models banned entirely. Whether it's because of "hurts people's health" or some other moral panic, it will kill this tech off.
gpt-oss isn't bad, but even models you cannot run are worth getting since you may be able to run them in the future.
I'm hedging against models being so nerfed they are useless. (This is unlikely, but drives are cheap and data is expensive.)
Guardrails are just adjustable parameters. The trick is finding the right ones and turning them off.
I forget the name, but there is at least one project dedicated to this.
Here's the thing: this is like 0days, the bar gets higher and higher on served models from OpenAI/Microsoft/Google. Sure, some have api access and you can allow spicier content but a better answer is: "tl;dr: not reading that."
I'm tired of having to step around megacorp's opinions on how I should be running my software, what questions to ask. gpt5.2 is HELLA paternalistic when you ask for deep dives on technical questions. Try asking something near to state of the art and see how far you get, most decisions come with huge warnings about how "this idea won't work" etc.
This deviates from my expectation, which is a fun word calculator to help me with back of the napkin calculations quickly so I can do other things. If that's not the core use case, fine. But that is a flag to me that says this will continue to be a moving target, which is the OPPOSITE of what a tool is.
There are some projects to reverse alignment and other things, and those people are doing god's work for sure. For me, I want to be sure I have the ability to run these things on my own because I don't trust that this won't mutate into something not useful in my way on the long term.