Claims › Machine judge › C023
C023
In their own words · Bai et al. 2022 · §1 Introduction · used
…We also aim to develop methods that encode desirable AI behavior in a simple and transparent form, and that make it easier to understand and evaluate AI decision making.
“In this paper we develop a method we refer to as Constitutional AI (CAI), depicted in Figure 1, and use it to train a non-evasive and relatively harmless AI assistant, without any human feedback labels for harms. The method therefore improves upon, and partially replaces reinforcement learning from human feedback […] .”
The new assistant ‘RL-CAI’ is preferred by crowdworkers over those trained with previously collected [Bai et al., 2022, Ganguli et al., 2022] human feedback labels for harmfulness.…
Harmlessness was trained by having the model criticise and revise its own answers against a written list of principles, instead of having people label every harmful output. Human judgement moves out of the labels and into the rules.
The five layers, for this claim
The same core in the infographic, with the claims it touches.
Relations
No relation to another claim has been declared yet.
Notation
- Claim typeWhat kind of statement it is, so that the right test can be applied. Proposed until checked by hand.
- D1.9 · procedural-instructional
- FunctionWhat a claim does for the programme: a conceptual tool, a working method, corrective evidence, a technical reference.
- working method
- StanceHow the programme holds the claim: assumed, used, or contested.
- used
- Used in
- /lab/2026-07-21-what-counts-as-a-verdict/
- Cite this
- claudiocammarano.com/en/claims/c023/