gexijin / chatlize Goto Github PK

Hi. Nice project, runs great.

A minor but interesting issue: sometimes the agent is forgeting and overwriting instructions.

Playing with the demo 'Scatering Plot', it goes perfect until the 'Add a linear trend line' instruction. In both cases A and B, the agent uses a regplot with its own new axis labels, overwriting the previous intruction to rename labels as 'Petal Prop (cm)'.
Later, in the final 'Change the color of data points based on Species' happens something similar. In case A the agent change the colors and remembers the true labels, but forget about the trend line and the pearson. In case B the agent remembers the trend and Pearson, but forget the labels and to change the colors!

I know this specific situation is very minor, but shows that maybe the agent's logic will improve including an instruction to keep the previous instructions in mind.

Hope being of help, regards

Andrés FRP.