What are your thoughts on AI?
Me: List five British actors aged in their 40s whose surnames start with S.
ChatGPT: Here are five British actors in their 40s whose surnames start with "S":
Your phrasing needs to be more explicit - I tweaked it to be:
list 5 british actors with a surname starting with the letter s and aged between 40 to 49
and it gave an improved (but still imperfect) list:
I suspect that it needs to be refined further as it is not really picking up the surname value. I tweaked "starting with" to be "beginning" and it improved again.
It does show that it is a mix of us assuming the way we phrase it will match its understanding. It is improving however at an exponentional rate.
Your phrasing needs to be more explicit - I tweaked it to be:
list 5 british actors with a surname starting with the letter s and aged between 40 to 49
and it gave an improved (but still imperfect) list:
I suspect that it needs to be refined further as it is not really picking up the surname value. I tweaked "starting with" to be "beginning" and it improved again.
It does show that it is a mix of us assuming the way we phrase it will match its understanding. It is improving however at an exponentional rate.
I'm quite experienced in using this particular chatbot model
What the hell do you use it for.......IRL?......based on your own assessment of its capabilities? Genuine question - I am interested!
Your phrasing needs to be
Respectfully, my phrasing doesn't "need to be" anything - although I do appreciate that you were trying to help.
I'm quite experienced in using this particular chatbot model, and I know that asking such a question in this way is pretty futile.
I posted the exchange to demonstrate how, when it’s asked questions using colloquial English, it can regularly fail to answer them correctly.
Not only that, but it will also keep on making similar errors when asked to try again, even when the user has stressed that all of the criteria must be met (the same criteria that it keeps relaying back to that user each time, as if confirming that it's understood them).
The example I've used is one that enables us to see, pretty easily, that what it said is wrong. Bigger issues can obviously arise if errors are less clear to us.
In my view, we shouldn't need to phrase our questions in very specific and rigid ways (similarly to how, say, computer programmers / coders might structure queries). That, to me, would defeat the object of the exercise.
Of course, there are different and much more advanced models than this, which will return better results. Capabilities will obviously also improve over time, including in the "free to use" versions.
In the meantime, though, I feel it's always worth treating replies from chatbots with a high level of caution.
Olivia Coleman is a stage name......her actual name is Sarah Sinclair.
