Feature #3600
Fires on “ an” (not labeled yet: read the examples)
- Fires on
- 0.56% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 74%
- story 16%
- chat 6%
- python 3%
- worksheet 1%
Makes the model say
_ angry work proudairTogether important acc
Makes it not say
hes wind So could not bring dr toy
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 4.92ute, lsp<|system_end|><|system_start|>You are an AI agent running in
-
agent 4.79ery, memory<|system_end|><|system_start|>You are an AI agent running in
-
story 4.71is 257.<|assistant_end|><|bos|>Once upon a time there was an old man. He
-
agent 4.67search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
story 4.63small town. All the kids were so happy to go. There was an ancient tree in the
-
agent 4.6search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 4.59search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 4.59search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 4.58search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 4.58search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 4.57search, write, execute<|system_end|><|system_start|>You are an AI agent running in
-
agent 4.57search, write, execute<|system_end|><|system_start|>You are an AI agent running in