Skip to content
mini-lab
Log in Sign up
Features of prelude-2

Feature #364

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

Fires on
1.27% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 77%
  • story 18%
  • chat 3%
  • python 1%
  • worksheet 1%

Makes the model say

rd can'ge should:`.ap

Makes it not say

good tell returns they says not next and

Strongest examples

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

  1. story 4.95
    big box in the middle of the room. His mum said, "We can unpack the
  2. agent 4.71
    , subagent, webfetch
  3. agent 4.71
    , subagent, webfetch
  4. agent 4.71
    , subagent, webfetch
  5. agent 4.71
    , subagent, webfetch
  6. agent 4.71
    , subagent, webfetch
  7. agent 4.71
    , subagent, webfetch
  8. agent 4.69
    , question, read, shell, skill, subagent, webfetch
  9. agent 4.68
    , question, read, shell, skill, subagent, webfetch
  10. agent 4.65
    , question, read, shell, skill, subagent, webfetch
  11. agent 4.64
    , question, read, shell, skill, subagent, webfetch
  12. agent 4.62
    , question, read, shell, skill, subagent, webfetch