Skip to content
mini-lab
Features of prelude-2

Feature #1713

Fires on “ are” (not labeled yet: read the examples)

Fires on
0.53% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 70%
  • story 25%
  • chat 4%
  • worksheet 0%
  • python 0%

Makes the model say

very good both brave alsoal am

Makes it not say

wind[- morningcaselist doornumsze

Strongest examples

On text the dictionary never trained on. The underlined token is where it fires most.

  1. agent 4.12
    bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
  2. agent 4.12
    bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
  3. agent 4.12
    bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
  4. agent 4.12
    bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
  5. agent 4.12
    bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
  6. agent 4.12
    bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
  7. agent 4.12
    bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
  8. story 4.09
    , "You are very talented!
  9. agent 4.08
    bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
  10. agent 4.08
    bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
  11. agent 4.08
    bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
  12. agent 4.06
    bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running