Generative AI & LLMs
Alignment tax
Extra costs or reduced performance that can come from making AI follow desired rules and behavior.
Example
A training approach improves safety performance while increasing development cost.
Why people use it
It helps compare the resources or capability changes involved in improving AI behavior.
What you'll hear
“What did that safety improvement cost?”
What this means for you
Compare specific benefits and costs rather than assuming a universal penalty.
Can you control it?
No
No direct control. This describes a wider issue, concept or result rather than something you can simply switch on or off in a tool.
Common questions
- Does alignment always reduce useful capability?
- No. Tradeoffs vary, and some alignment work can improve usefulness.
- Can the cost be development time rather than worse answers?
- Yes. Extra testing, training or operating work can be part of the tradeoff.
- Is there one standard number for the tax?
- No. The result depends on which benefits and costs are being compared.