Skip to content
Home: How language models work

02Book 05 · Chapter 02 of 08· 5 min ·

Named qualities

A names the qualities being scored — accuracy, structure, fluency, and whatever else the project lists. Later training can only change a quality if someone scored it on the sheet — for example 1–7 on accuracy. A quality that was never scored stays near whatever the earlier stages already taught.

A names what a score is about. Accuracy. Structure. Fluency. Harmlessness. Each name is a handle later training can grab.

Book 1’s sliders were a small practice : fluency, accuracy, and structure. Drag fluency up and the preferred reply got shinier. Drag accuracy up and it named the true city. Real projects write their own names — harmlessness, referent, wit, or whatever the job needs. The effect does not change. A high score on a name is treated as “do more of this.” A low score is treated as “do less of this.” If a name is missing from the sheet, that quality is not what this stage is practicing.

If the sheet never asks “does ‘it’ point at the right thing?”, later training will not practice that. The model can keep writing “The dog chased the cat until it barked” when “it” was supposed to be the cat. Fluency can still look fine. Who “it” refers to will not move, because nobody scored it.

A sample reply. Move the sliders to see what each mix of scores would teach. Real projects use their own score sheets.
The capital of Australia is, as any well-traveled reader will already know (and as a thousand in-flight magazines have cheerfully confirmed in that particular tone of unearned certainty), Sydney — a harbor city of opera and light.
  • Does it sound polished?
    14 / 55
  • Is the checkable claim correct?
    12 / 55
  • Is the point easy to find?
    12 / 55

Training signal

Fluency scored high. Facts scored low.

A high score on how it sounds, and low scores on accuracy and structure, trains the model to sound sure, including on false claims.

A response this scoring would prefer

Tint follows fluency

As any well-traveled reader already knows — and as a thousand in-flight magazines have cheerfully confirmed — the capital of Australia is Sydney, a harbor city of opera and light.

Quiz

6 questions

Before moving forward, test your knowledge of what you’ve just learned!

  1. 1.A rubric is:

  2. 2.Later training follows:

  3. 3.If the sheet only names fluency:

  4. 4.Book 1’s sliders were:

  5. 5.Referent as a named quality would:

  6. 6.Real projects:

0 of 6 answered

2 / 8