You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It could be beneficial, if human feedback could be persisted in long term memory.
I didn't fully think this through but there could be many instances where the operator gives feedback because the model misunderstood the task. After giving feedback, the context and feedback could be stored in long term memory to help AutoGPT make the right decision immediately next time, without the need to provide feedback again.
Note that this is most beneficial, if memory is persistent - as far as I understood, that is not the case right now(?)
Examples 🌈
I read in another issue that AutoGPT wanted to clone its own repo github.com/Torantulino/Auto-GPT but ended up finding and cloning GPT-2.
Human feedback could be: When I talk about AutoGPT i mean github.com/Torantulino/Auto-GPT.
On a future run, AutoGPt would know, that GPT-2 is not AutoGPT and would not make the same mistake again.
Motivation 🔦
autonomous learning of AutoGPT or at least learning from mistakes.
The text was updated successfully, but these errors were encountered:
This issue has automatically been marked as stale because it has not had any activity in the last 50 days. You can unstale it by commenting or removing the label. Otherwise, this issue will be closed in 10 days.
Duplicates
Summary 💡
It could be beneficial, if human feedback could be persisted in long term memory.
I didn't fully think this through but there could be many instances where the operator gives feedback because the model misunderstood the task. After giving feedback, the context and feedback could be stored in long term memory to help AutoGPT make the right decision immediately next time, without the need to provide feedback again.
Note that this is most beneficial, if memory is persistent - as far as I understood, that is not the case right now(?)
Examples 🌈
I read in another issue that AutoGPT wanted to clone its own repo github.com/Torantulino/Auto-GPT but ended up finding and cloning GPT-2.
Human feedback could be: When I talk about AutoGPT i mean github.com/Torantulino/Auto-GPT.
On a future run, AutoGPt would know, that GPT-2 is not AutoGPT and would not make the same mistake again.
Motivation 🔦
autonomous learning of AutoGPT or at least learning from mistakes.
The text was updated successfully, but these errors were encountered: