liveBook · Manning

Dual-use technology

As explained in How Large Language Models Work

Dual-use technology is technology that can support both beneficial or peaceful applications and harmful or adversarial applications.

Dual-use technology can support peaceful goals and harmful ones. With large language models, improved control, safety, or capability can help beneficial users while also helping adversaries. section 9.2.2

The backdoor analogy

An encryption backdoor intended only for trusted users can become an attacker’s most valuable target. Similarly, controls designed to make an LLM more useful may create capabilities that increase risk for everyone. section 9.2.2

Assess dual-use risk by asking:

  • Who would be motivated to cause harm, and why? section 9.2.2
  • What resources or actions would the harm require? section 9.2.2
  • Which existing barriers prevent the harm, and can modern technology bypass or strengthen them? section 9.2.2
  • How do surrounding systems enable or block success and risk? section 9.2.2

How How Large Language Models Work covers it

How deep it goes
Defines the risk and supplies threat-model questions, but stops short of specifying particular harms, mitigations, or technical procedures.
Examples
Uses an encryption backdoor analogy and questions adversary motivation, required resources, barriers, and surrounding systems.
What you can do after
Evaluate LLM-related risks by identifying adversaries, motivations, barriers, circumvention paths, and enabling systems.
How it connects
Connects dual-use technology with alignment, encryption, threat models, barriers, and LLMs.
Cover of How Large Language Models Work
liveBookmanning.com
Read How Large Language Models Work on liveBook Get the book at manning.com

This explainer was generated by Manning's knowledge graph from the text of How Large Language Models Work. The section links point to the parts of the book it draws on.