x

LESSWRONG

LW

Georgi Shopov — LessWrong

Georgi Shopov

Georgi Shopov

Message

Independent, self-taught AI safety researcher. I've spent the last few years mostly alone stress-testing language models, building the Pandora Benchmark, and trying to turn a suspicious amount of controlled madness into a usable theories of runtime alignment, adversarial evaluation, and model behaviour under pressure.

2

20d

Georgi Shopov

Independent, self-taught AI safety researcher. I've spent the last few years mostly alone stress-testing language models, building the Pandora Benchmark, and trying to turn a suspicious amount of controlled madness into a usable theories of runtime alignment, adversarial evaluation, and model behaviour under pressure.

Runtime Objective-Orientation as a Theory of AI Alignment

Alignment Is Not What a Model Has. It Is What a Model Becomes Oriented Toward Important note: I am new around here. And I want to take you on a short, partly technical and slightly intellectual journey into the field of AI and the Alignment problem. I did my best...

Runtime Objective-Orientation as a Theory of AI Alignment

Alignment Is Not What a Model Has. It Is What a Model Becomes Oriented Toward The problem A lot of alignment discourse treats alignment as something a model has: a property shaped during training, checked during evaluation, and then preserved or degraded during deployment. I think this framing is incomplete....