Now what will happen, will they make GPT-4 for everyone by default or something else they are gonna come up with.
GPT-3.5 was okay for my Scientific and mathematic answering. But now they have removed it. And GPT-3 lot of times does not what it is even answering. Even further telling about his answer as wrong, it keep running hoops.
Edit: When I typed this I was not aware that the 3-May-2023 update got description edited.
Default and Legacy
the legacy one will be removed, default will stay
Thats my understanding
( assuming you are talking about the last update notes of chatGPT, I didnt see anything about whole 3.5 being deprecated )
I first saw in chatGPT plus the note next to the model before there were any update notes.
Then I saw the update notes the day it was put out and it always mentioned specifically legacy model being deprecated
(they also messed up the date of update on bottom of chatGPT which they later fixed)
We’ve made several updates to ChatGPT! Here’s what’s new:
You can now turn off chat history and export your data from the ChatGPT settings. Conversations that are started when chat history is disabled won’t be used to train and improve our models, and won’t appear in the history sidebar.
We are deprecating the Legacy (GPT-3.5) model on May 10th. Users will be able to continue their existing conversations with this model, but new messages will use the default model.
I think it was the one that was available in chatGPT before there was that switch to turbo.
From description in chatGPT:
Legacy (GPT-3.5) [Deprecating Soon]
This is the previous Plus model. It will be deprecated on May 10th. Please switch to Default (GPT 3.5) or GPT-4.
I rather they use the resources for the actual 3.5 turbo and GPT4
They have to discontinue outdated less efficient models to make space for the new ones
They all take resources to run
it also shows lower speed in the description of the model
Which might mean it took more resources to run than the Default 3.5
I hope they bring feature of persona. Default talks too much for my answers. I want one that directly shows maths, writes every step, since I can too understand where that numbers came from without tell much about context.
An ability that the Legacy GPT-3.5 model had was the ability to number it’s responses. The “default” GPT-3.5 doesn’t have this capability, but the ability is retained in GPT-4 from my experiments. Roughly what I was doing was asking ChatGPT to number each of it’s responses with “01” then “02” etc. I then could tell it to refer to the response number 12 for example and elaborate on that answer even if I was already on response 23. When not using this approach I noticed ChatGPT would be unable to elaborate on a previous response because there was no context. Anyway, ChatGPT should build that into the model by default in my opinion.
I just read the update and was concerned too, but the answers on this post have put them to rest. On another note, I learnt that “to deprecate” is a verb you can correctly use in the context of IT for things like this. I was only aware of the regular sense so I thought the update had a spelling mistake lol:
Haha, I get the feeling some techie was tryng to use a word and they didn’t quite know what it meant, and then others thought it sounded intelligent so just started using it too and it caught on. May not of happened that way, but it’s funny to think about.