Skip to content
mini-lab
Features of prelude-2

Feature #3438

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

Fires on
0.51% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 79%
  • story 11%
  • python 7%
  • chat 3%
  • worksheet 0%

Makes the model say

" python its welcome dog} from this

Makes it not say

slppatternouillioush

Strongest examples

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

  1. agent 3.75
    ent, webfetch, websearch, write, exec
  2. agent 3.75
    ent, webfetch, websearch, write, exec
  3. agent 3.73
    ent, webfetch, websearch, write, exec
  4. agent 3.73
    ent, webfetch, websearch, write, exec
  5. agent 3.72
    ent, webfetch, websearch, write, exec
  6. agent 3.71
    ent, webfetch, websearch, write, exec
  7. agent 3.71
    ent, webfetch, websearch, write, exec
  8. agent 3.71
    ent, webfetch, websearch, write, exec
  9. agent 3.7
    ent, webfetch, websearch, write, exec
  10. agent 3.7
    ent, webfetch, websearch, write, exec
  11. agent 3.69
    ent, webfetch, websearch, write, exec
  12. agent 3.68
    ent, webfetch, websearch, write, exec