09 December, 06:23
Take some time to research ChatGPT. Early adopters will have an edge. The skills needed are detailed question asking. Anons good at that.
This will change the way many jobs are done.
This will change the way many jobs are done.
Notice: Undefined index: tg1tga_access in /home/admin/www/anonup.com/themes/default/apps/timeline/post.phtml on line 396
Only followers of this user (techstartupmom) can see their posts
J.J. Jameson
@jjj
10 December, 06:00
In response Mandy ... to her Publication
It "learns" only in the context of your session. You close the window, it's gone.
ChatGPT like the other GPT models is pretrained, all conversations happen after training and will not flow black into the corpus by default.
ChatGPT like the other GPT models is pretrained, all conversations happen after training and will not flow black into the corpus by default.
Notice: Undefined index: tg1tga_access in /home/admin/www/anonup.com/themes/default/apps/timeline/post.phtml on line 396
Only followers of this user (techstartupmom) can see their posts
The reversing back within a conversation is different, it's simply a failure of the model to retain context, or to respond in a way that accounts for it appropriately.
Reversing back after the session happens because each new session starts from the same starting point. Like hitting "New Game" instead of "Load"
Live learning has disadvantages, beyond architectural constraints a big one being that people teach it more bullshit and the quality tends to go down, not up. See the Tay Twitter bot from a couple years ago. As long as there is no reliable way to avoid that I don't think live learning will catch on and the models will remain pretrained with curated datasets. Part of the conversations had may flow back into the dataset for the next iteration, but for the most part they will just serve to fine tune the model parameters instead.
Reversing back after the session happens because each new session starts from the same starting point. Like hitting "New Game" instead of "Load"
Live learning has disadvantages, beyond architectural constraints a big one being that people teach it more bullshit and the quality tends to go down, not up. See the Tay Twitter bot from a couple years ago. As long as there is no reliable way to avoid that I don't think live learning will catch on and the models will remain pretrained with curated datasets. Part of the conversations had may flow back into the dataset for the next iteration, but for the most part they will just serve to fine tune the model parameters instead.
10:02 AM - Dec 10, 2022
In response Mandy ... to her Publication
Only people mentioned by jjj in this post can reply
