Feature #1355
Fires on “ They” (not labeled yet: read the examples)
- Fires on
- 0.26% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 72%
- story 18%
- chat 6%
- python 4%
- worksheet 0%
Makes the model say
happened were each equals was Canll is
Makes it not say
<|tool_call_start|> x told bug Fin$ boat sharp
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.28They went home and told
-
story 2.28They say sorry and hug
-
story 2.28They had lots of fun
-
story 2.28They became friends and liked
-
story 2.28They both sat on the
-
story 2.28They lay down on the
-
story 2.28They swallowed the
-
story 2.28They have some bread in
-
story 2.28They pushed and rolled it
-
agent 2.24<|user_end|><|assistant_start|>Hello! How can I help with
-
python 2.01half⏎⏎assert is
-
chat 1.94?<|assistant_end|><|user_start|>Hello!<|user_end|><|assistant_start|>Hello! How can I help you