Let's Stop Pretending–ChatGPT Isn't That Smart
barrons.com
barrons.com
The iPhone and iPad gave us Star Trek like touch screens. ChatGPT has given us god like HAL interfaces.
1) IRC bots could not write novel text, just pre-planned responses. They cannot respond to everyday conversational language.
2) IRC bots do not have anywhere near as much information on hand.
3) IRC bots cannot hold context over multiple conversational exchanges.
4) IRC bots cannot even approximate fooling a human into thinking it was another human on the other end.
5) IRC bots absolutely can not improvise code from the context of your code base, let alone take plain language instruction.
Regardless of the above, this is the very beginning of the tech. I imagine if you were alive in the 80s you'd ask why the internet is any better than your local dial up BBS.
Even if it manages to fulfill the "HAL" interface, that isn't necessarily all that interesting. We've had Siri for a while, and it's regarded as a minor convenience, not a revolution. Certainly nobody is going to trust it with the pod bay doors.
There really could be something more there. But ChatGPT isn't that smart. It does things that don't need to be done; any text that it can write usually doesn't need to be written at all.
Still... it feels so much like at any moment it might accidentally start writing something worthwhile. Nobody knows what magic is missing. Arguably, it's missing all of the magic, and despite being a good party trick, it's not actually all that much closer than ELIZA was.