Ive noticed it quit giving as detailed answers and as thorough. It's also refused to do more complex programming where it used to accept those questions.
Being artificially limited by OpenAI can still be done without it getting "worse". But it effectively is worse for us users.
The OG models were trained on real world, human generated content (for the most part at least). Starting in 2022 the cost of automatic generating "human sounding enough" text has gone to such low depths that I expect it to be pretty much impossible to avoid training any model on text already generated by a LLM.
What will the result of this feedback loop be, I can't tell. It will probably be just an even more generic, corporate speak, bland sounding bla bla bla than we get today, and the level of hallucinations may get even worse.
In a way it makes me happy to imagine that the most dangerous tech humanity ever invented may itself be it's own main obstacle to future refinements.
it could just save every answer its given and scan text for it. If they're a match it could just not index it, right?
Who, in turn, will use LLMs to generate that text.
Seems like we need "model transparency" and log implementers to flag drift, a la RFC 9162 / Certificate Transparency.
It has a very basic prompt on top of the existing models. There is no additional fine tuning involved.
Moderation
Temperature / top-p
System prompt
Some other internal system we aren’t aware of
The release notes are producty and not very technical so it is difficult to tell what actually changed.
It didn't. People were just noticing LLMs still have a long way to go after using them more.
I was shocked going through the thread that all of the popular comments were confirmations that it did, in fact, get MUCH worse.
It's nice to see from OpenAI that it didn't...