I posted this in another thread. I did play around with it but was exploring how well it did in the Turing test.
Quote:
Originally Posted by Edward64
I'll take back the easily exceeded comment.
ChatGPT ChatGPT: Optimizing Language Models for Dialogue has been in the news recently about how great it was etc. So I played with it.
If the test is the Turing Test, it fails. I believe I can easily discern if I'm talking to a computer vs human.
If the test is can it do or "understand" much more than a helpdesk chatbot or Siri, yes it can.
FWIW some observations:
- I asked it to do haikus for the Razorbacks and for the Tide. The haikus were fairly generic (e.g. no mention of WPS, Hogs, Tide, Elephant) and both were similar to each other
- I also asked it for its opinion about a subject matter. It said it was not "programmed" to give opinions.
- I asked about something in 2022 and it said it was only "programmed" for up to 2021. This one is kinda strange. I've been around AI that supposedly scrapes news/research content and stores it in its internal-whatever-databases. I'm sure there's a reason but sounds very limiting IMO to not have ChatGPT perpetually scrape for news/research content
- And the most telling on why it failed the Turing test, its responses didn't seem to flow naturally. You could tell you weren't having a conversation with human
Overall, pretty interesting and looking forward to see how it progresses. But not yet there in the Turing test. |
Quote:
EDIT: article on use of ChatGPT to write essays. My internal debate now is whether to let my daughter know this exists!
|