Skip to content
mini-lab
Features of prelude-2

Feature #2984

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

Fires on
0.94% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 58%
  • story 33%
  • chat 6%
  • worksheet 3%
  • python 1%

Makes the model say

brave enough and even ataverage should share

Makes it not say

=:case + noises<|assistant_end|>", "!"

Strongest examples

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

  1. agent 5.2
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in
  2. agent 5.18
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in
  3. agent 5.14
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in
  4. agent 5.13
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in
  5. agent 5.13
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in
  6. agent 5.13
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in
  7. agent 5.11
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in
  8. agent 5.09
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in
  9. agent 5.09
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in
  10. agent 5.09
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in
  11. agent 5.09
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in
  12. agent 5.08
    search, write, execute<|system_end|><|system_start|>You are an AI agent running in