What if LLM design ignored accuracy?

So a lot of the conversation around LLMs seems to centre on accuracy (aka propensity to hallucinate) and by extension, risk and security.

I wondered what would happen if you just said who cares about accuracy?

What if the design challenge focused more on stuff GPTs are good at (like spitting out lots of options) and less on the risks caused by inaccurate information?

For example, you might have a workshop canvas or a form you’d been meaning to complete but for whatever reason you’re stuck. Turns out GPT is really good at showing you examples.

You still need to think it through and make your own content, but this ‘boost’ is really valuable:-)