Roman Malov comments on Stephen McAleese’s Shortform

[Roman Malov]

the “one try” framing ignores that iterative safety improvements have empirically worked (GPT-1→GPT-5).

That’s because we aren’t in the superintelligent regime yet.