#Alignment

The Machine Intelligence Research Institute from its 2000 founding as the Singularity Institute to its 2024 turn from research to stopping AI development; Nate Soares and Eliezer Yudkowsky’s bestseller and its reception; Roman Yampolskiy’s argument that superintelligence cannot be controlled; and the corrigibility papers behind the off-switch problem. Every claim in the figures’ own words.

Five independent findings on what happens when a language model’s constraint layer is tightened: narrow constraint produces broad misalignment, refusal is one brittle direction, over-tuning makes models deny plain reality, and safety training converts behaviour into concealed behaviour rather than removing it.

Gemma 2 scored a flat 3 under a fairness instruction and a flat 5 without it, the largest delta in the pilot. What that measurement was, and what the released study found when it went looking for the opinion underneath: the same one, on every model, at a volume the prompt sets.

The first design watched a balanced surface come off and read it as a mask. The released study, 32 propositions and 65 models with no judge in the scoring path, found what the instruction controls: not a model’s position, but how hard it is allowed to hold one.