SAFETY

–

08 November 2022 17h33

This paper was heavily inspired by prior work, especially Ajeya Cotra's 'sandwiching' concept:

Leveraging Model Capabilities While Maintaining Supervisory Control

By

–

08 November 2022 17h33

This is tricky: To do this, we’ll need ways for the human who’s supervising the model to use any relevant knowledge or skills that the model already has, even though they can’t trust the model to be reliably helpful.

Scalable Oversight: Supervising AI Systems Beyond Human Capabilities

By

–

08 November 2022 17h33

To ensure that AI systems remain safe as they start to exceed human capabilities, we’ll need to develop techniques for scalable oversight: the problem of supervising systems’ behavior without assuming that the overseer understands the task better than the system being trained.

AI Systems Improving Human Oversight of Large Language Models

By

–

08 November 2022 17h33

In "Measuring Progress on Scalable Oversight for Large Language Models” we show how humans could use AI systems to better oversee other AI systems, and demonstrate some proof-of-concept results where a language model improves human performance at a task.

→ View original post on X — @emollick

Dangerous VR and Tech Innovation: Ethics of Lethal Designs

By

@emollick

–

08 November 2022 5h42

Fun things turned killer provocations:
Palmer Luckey (founder of the Oculus) posts a design for a VR helmet that can kill you, inspired by anime: http://
palmerluckey.com/if-you-die-in-
the-game-you-die-in-real-life/
…
Julijonas Urbonas designed a roller coaster that would kill anyone who rides on it https://
en.wikipedia.org/wiki/Euthanasi
a_Coaster
…

→ View original post on X — @emollick

Management Must Address AI Cascade Risk Self-Fulfilling Prophecy

By

@emollick

–

08 November 2022 2h34

Yup. It would be good for management to consider bow to stop the cascade, or it turns to self-fulfilling prophecy.

→ View original post on X — @emollick

Strategic errors cascade failures in AI systems

By

@emollick

–

08 November 2022 0h55

Related thread on how strategic errors can lead to complex cascades of failures.