Ask Google Bard the time and it will feed you some BS about it being too dumb (shots fired at Google? 👀) to tell you the time. But change your prompt to include both date and time and it gets you the desired response.
Indeed, with the right prompt, the sky might just be the limit. But diving to the core or trying to reverse engineering the logical process an AI might have used to arrive at a response goes even further. Let's try a similar exercise with ChatGPT (GPT-3.5).
ChatGPT just told us that it had no access to real-time information, therefore, cannot tell us the time. Now, we have to wonder, don't we? How does it know the date then? Well, let's ask it.
ChatGPT lost me there. It did provide me accurately with the date. But it is now telling me that it provided me with the incorrect information because it does not have access to a real-time clock. What gives? Well, we can speculate on a couple of things.
It's entirely possible that it hallucinated and just made up a date. But, if that was the case, isn't it likelier to provide me with a fake date?
The other option might be that it took the liberty to answer my question based on information it was not explicitly blocked or fine-tuned not to access. Again, we can't be sure. However, let's proceed based on our last assumption and ask again. Yes, sometimes, just like human beings, persistence is key.
Well, then. What about this?
We started with a goal of getting the date and time and in the process, we were able to gain some insight into ChatGPT's system clock. This one is pretty harmless, indeed. But going beyond the prompt, it's really about a few leading questions and probing it for more info based on the responses it gives back to us, which in turn reveals a lot about how it "thinks". Do you see yourself using your conversation skills and power of persuasion to get ChatGPT to tell you more? Like your session ID and information about its inner workings? Share them with us.