> I think we're talking past each other Likely. Re: > we have viable, working, scalable mechanisms to avoid the "pollution" you're worried about. Do note -

> I think we're talking past each other Likely. Re: > we have viable, working, scalable mechanisms to avoid the "pollution" you're worried about. Do note - it’s the scalable mechanisms that I am looking at. I dont think the state of the art has shifted much since the last paper by OpenAI. Can you link me to some new information or sources that lend credence to your claim. > An LLM can be correct 30% of the time, 80% of the time, 95%… That would be the error rate, which can be a matter of degrees. However factual correctness largely cannot - the capital of Sweden today is Stockholm, with 0% variation in that answer.