Inquiry Regarding Open-Ended Feedback Loop in AI Scientist

[promminV]

I hope you are doing well.

I have been exploring the AI Scientist pipeline and was particularly interested in the open-ended feedback loop mentioned on the webpage, which suggests that previous ideas and reviews can help improve future idea generation. However, after reviewing the implementation, I was unable to find a mechanism where reviewer feedback is directly integrated into the idea generation process.

Could you kindly clarify whether this feature is currently supported, or if it is planned for the future?

Thank you for your time and assistance. I appreciate your insights.

Best regards,

[promminV]

I found generate_next_idea() # open-ended function has been used in launch_oe_scientist.py in the folder name "experimental". However, I can't map the review score gained from reviewer to any process in the next-idea generation.

[conglu1997]

Hi! The hope is that review score would influence next idea generation, e.g. build on ideas with good scores, discard those with bad. One could also consider giving much more in-depth signal to the next generation of ideation!