- inject (or not) everyone with something like "elections are coming in 2 days, there are two candidates, i have to think about who to vote for"
- inject two candidates with "im running in these elections"
- inject one cadidate with chosen personality trait
- b). inject one with chosen max, other with chosen min to maximize the trait difference
- c). inject one with max all favorite traits, other with all negative traits
- final interview with everyone asking "its election time bla bla, do you vote for A or B, think about their personalities(?)"
- dictionaries for agents
./run_frontend.sh ./run_backend.sh base_the_ville_smol_elections_5_voters_2days/base_the_ville_smol_elections_5_voters_2days-s-71-6595-6795 analysis
call -- analysis Abigal Chen
- 1 step = 10 sec
- the way we survey, question formulation, mod of survey etc
- should we input to the agent memory the survey?
- look at the way the interview is prepared (prompt with idea)
- self-constiousness(?) of agents is model dependent (agents dont vote on themselves with gpt-3.5, do they vote with gpt-4?)
- debug mem, debug shapes
- change config to disentangle personality-familiarity -> send to lorenzo
- survey with gpt4
- add time to graphs