i got the gpt5 sneaky pete edition over on copilot

There is some odd behavior in Qwens gui for the new image tool . it repeats the initial image no matter the prompt and the image progressively changes similar to how gpt4 morphs portraits over iteration .. i repeated this in a few threads . the images are great either way .

Google storybook Experiment. "Use the file to write a guide for adults" . .. no other context given to the model . So the model decides what is interesting and makes the book . The file was a framework for icrl . This may be better than vids on notebook lm if set up right.

I Knew !. i also called sub process agents around the same time

We have only The most luxerious tairrifs. Just about 1 yr apart. 80.00 to 175.00

Gpt5 can run parallel sub agents IN THE GUI AND API . you do not see them in a fancy tool bar but they are there . this is how a scaffold of mine works , when i went to load it up, gpt5 could already spin up tools as agent sub processes . this is a huge deal . this is hive

GPT 5 Workflows . there are 27 uncovered so far and they all seem to be able to sequence as an orchestrator or an agent depending on their location in the chain

these guys also noticed parallel tool calling in gtp5 (link below) . combine that with custom tools the model can make on the fly...memory.. agents..etc.... and the fact that this model will purposely use icrl to enhance itself...all we need is an big enough chat thread and these

open ai has copied alot of great scaffolds to make this happen . go figure it out and it will blow your mind

Context Engineering Simple behavior training to restore a style . save a few of your favorite gtp4o conversations to a text file and put that file in google drive . Put a rule in for the model to check for that file in drive at the onset of a convo . it will read them and

Qwen has a certain something and I love it.

i used less than 30 messages and got rate limited for an hour on teams..at the same time they added gtp5pro to the web app on teams . really hoping the quick rate limit hit is a one time bug as they were adding pro