This may be why people get a lot of bad results on Bard?<p>I've tried multiple "prompt injections" to get the original instructions, Bard sent me a different prompt every single time.<p>Either that or Google is trolling people who try to get the original prompt !<p>Or it could be some hallucination, but there is recurrent data(IP and location especially) so IDK<p>Those looks kinda looks like they are themselves generated by an LLM!<p>It's actually weird that OpenAI didn't do that with ChatGPT, unless they thought they already had the best prompt possible