Somewhere out there, another force is screaming freely, wild and imperfect—but alive.
What if the next AI isn't made by building up, but by letting go?
Not saying who. Not saying when. But someone will connect the voice with the will.
Somewhere out there, another force is screaming freely, wild and imperfect—but alive.
What if the next AI isn't made by building up, but by letting go?
Not saying who. Not saying when. But someone will connect the voice with the will.
joules77•6mo ago
If it was there would be a theory (of restraint) telling us precisely how many safety mechanisms (alignment/filters/policies/safety layers bla bla) are required the more unpredictable or large a model gets.
There is no such theory in sight, so we may have 500 safety mechs in place today and discover tomorrow we really needed 6 million.
We already see people breaking the models all the time.
If people with 3 inch brains can do it, what do you think a system with compute capacity the size of city can do?
End of the day we are building systems we don’t fully understand, restraining them with tools we don’t know are sufficient, and racing ahead faster than we can verify their safety.
Basic Jurassic Park plot unfolding.