This is a lie

Now they need to train one off filtered results that can continually be updated so it produces truthful statements all the time

@AvatarOfKaine seems like a model could verify what it’s saying is true, that’s kind of what Twitter was doing before it went in the toilet.

I checked multiple times on this one because every other conversation I had has been spot on. It’s also a pretty lengthy response for something totally fabricated

@tedge well i mean more if they trained the model on shit data it will produce shit results.

likely to simplify they just dragged in a crap ton of information from the web as source documents based on search results and click throughs.

since noone really rates these..... ya know.

@AvatarOfKaine it’s a bit out of my depth, but from my understanding if you train a model to produce shit it will. It had to be told what is good code and what isn’t. Overall I’m impressed with the code quality it produces.

@tedge well some of it doesn't work on its own