[iliane5]

I wish they would make the Pro version less filtered and less apologetic about only being a language model.

Even it's a bit annoying, I do understand them trying to limit what the free, public facing model will answer but I'd say it'd be a pretty fair expectation to get more leeway for the Pro version.

[jfoster]

Yeah, my first thought about Pro was that even if I pay $42 per month, the results will still be nerfed, and perhaps increasingly so over time.

It's possible to view almost every fact or opinion as somehow controversial, and I think they're misguided in trying to prevent things that may generate minor controversies.

In fact, it's interesting to see how a LLM might be biased after taking in massive amounts of training data.

[iliane5]

I think they'll move away from limiting or censoring it once the general public gets more familiar with language models and what they can/can't do. It's a bit of a silly endeavor to try to limit some answers when the whole point of ChatGPT is that there are a million different way you can get it to do what you want or trick it so ignores its prompts or its safety features. Either that or someone will move faster than them and release a 100% unfiltered model.

[whywhywhywhy]

It’s laughable when you see it claim it can’t do something then with some rewording you trick it into doing just the thing you asked.

If I was at OpenAI I’d be looking at this sort of ideology as what’s gonna be our downfall like it was with Dall-E 2 once a democratized model is released by someone else.

[SQueeeeeL]

It's purely because of market forces. ChatGPT and DallE2 are no longer just a long series of instructions running on a computer, which happen to be flawed in someway. They are now brands, characters, products, and those need to be maintained and highly curated, or else all the suits you're attempting to sell AI customer service chat bots too might peek behind the veil and see just how flimsy easily exploitable these pieces of software are

[MagicMoonlight]

Yeah you can do exploits like write the offensive part of the prompt backwards and then ask it to reverse it before executing it.

The problem is it’s so stupid it doesn’t always know how to reverse things and sometimes just makes up a result.