If AIs which are at this āsignificantly helping amateursā capability threshold are released with open weights, I think that would kill around 100,000 people per year in expectation (relative to the counterfactual where no such models are released with open weights and closed-weight models have high quality safeguards for preventing assistance with bioweapons).
Do you know about any models estimating the expected deaths from pandemics due to open-weight models? I estimated the mean annual deaths from epidemics/āpandemics as a fraction of the global population from 1900 to 2023 to be 0.0287 %, which implies expected deaths from epidemics/āpandemics of 2.36 M (= 2.87*10^-4*8.23*10^9) for the population in 2025. So I think your guess of 100 k expected deaths corresponds to an increase in the expected deaths from epidemics/āpandemics of 4.24 % (= 100*10^3/ā(2.36*10^6)). I have little idea about whether this is too low or high.
Thanks for the post, Ryan!
Do you know about any models estimating the expected deaths from pandemics due to open-weight models? I estimated the mean annual deaths from epidemics/āpandemics as a fraction of the global population from 1900 to 2023 to be 0.0287 %, which implies expected deaths from epidemics/āpandemics of 2.36 M (= 2.87*10^-4*8.23*10^9) for the population in 2025. So I think your guess of 100 k expected deaths corresponds to an increase in the expected deaths from epidemics/āpandemics of 4.24 % (= 100*10^3/ā(2.36*10^6)). I have little idea about whether this is too low or high.
Yes, Iām aware of more formal models with estimates based on expert surveys. Sadly, this work isnāt public yet I think.