Skip to content
mini-lab
Features of prelude-2

Feature #3991

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

Fires on
0.35% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 94%
  • story 4%
  • worksheet 1%
  • chat 1%
  • python 0%

Makes the model say

of both other the friendsstSorry laughed

Makes it not say

epinitialeretidaselfoo

Strongest examples

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

  1. agent 2.55
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
  2. agent 2.55
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
  3. agent 2.53
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
  4. agent 2.53
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
  5. agent 2.52
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
  6. agent 2.52
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
  7. agent 2.52
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
  8. agent 2.51
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
  9. agent 2.5
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
  10. agent 2.5
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
  11. agent 2.5
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
  12. agent 2.5
    arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode