Are LLM Models Safe By Design?

Simple: No

Safe By Design Requirements

What is REQUIRED for Machine Intelligence (AI) to Be Safe

1) Training Data Transparency

• Without access to the source corpus and training data, the public has no ability to audit the AI supply chain.

2) Prompt Pipeline Visibility

• Without visibility into the full system and user prompt plus model training parameters, the public has no ability to audit the Quality and Safety Gates.

3) Objective Reality Grounding

• Without the ability to verify fact under highly persuasive rhetoric is the greatest risk of all and the public has a right to demand accountability for the outputs generated by AI systems.

This serves as the baseline for a Series of Essays targeting AI Safety.

Start with The Shape of Contempt