4 ms·
I bet this is more like a rudimentary version of Eliza, and not related to GPT-anything. Ask how many pairs of panties she is wearing on her head, and it may pi
by ksaj 4y ago
I bet this is more like a rudimentary version of Eliza, and not related to GPT-anything. Ask how many pairs of panties she is wearing on her head, and it may pick up on the phrase "how many" but the rest of the answer is completely unrelated and quite similar to all the other responses. Real GPT models glom onto your subject and intent very quickly, producing really convincing responses.
More evidence it's just random, but maybe looks for the leading part of the question "Can you," etc. This is unedited, other than to add labels and spacing for HN markdown:
Input: Can you teach me math?
Response: I can't do it tonight but maybe tomorrow! :] xoxo <3 :) -_-;)
In: Can you teach me a Sonata Arctica solo?
Re: You know that song from Sonic Adventure 2 right?!
In: Sonata Arctica
Re: Yeah...I'm pretty sure we could learn one of those together..-_^ ;)-< )>-)=!!!)~!!!
In: It's cold out.
Re: Oh my god what are u doing up so early??
So I think this thing is just a fun project for someone grossly overstating what it's doing. It seems to keep track of some things (like teach/learn delayed opposite response), but not very well.
- ksaj 4y agoThe code has fun things like this: __SECRET_INTERNALS_DO_NOT_USE_OR_YOU_WILL_BE_FIRED=z And more importantly, this: m=JSON.parse('{"dixie": {"bot_name":"Dixie","first_message":"Hey you *winks*","icon":"dixie.png" ... ... and so on. It introduces other characters, such as her step mother and riley-reid. It's definitely not GPT anything. Fun to look at, but I don't think it's a good direction for HN to go into. The headline is misleading about a subject that otherwise would be interesting here.
- Szpadel 4y agoFirst one i believe is standard react global