← Back
Computation
Index: #52

AI Could Be Misused

The risk of AI being misused—whether through malicious intent or unintended consequences—necessitates robust safeguards and countermeasures.

Foundational Capabilities (2)

Decentralized Training

Maintain decentralized control over large neural network training, like SETIatHome/Wikipedia for large language models, to equalize access. However, this also introduces some AI proliferation and control related risks.
  • Prime Intellect
    Initiative

    • Jailbreak Resistance

    Develop techniques to ensure AI systems are resistant to "jailbreaking" or circumvention of built-in safety and control protocols.
  • Gray Swan
    Initiative