Feature #3099
Fires on “rite” (not labeled yet: read the examples)
- Fires on
- 0.93% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 68%
- story 17%
- chat 8%
- python 6%
- worksheet 1%
Makes the model say
justSorry perfect<|arg|>entss be then
Makes it not say
==8Help +two67dy
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.09_even that keeps only the even numbers.<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=lib
-
agent 3.07: it returns the product of a and b.<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=core
-
agent 3.05last_letter that returns the last letter of s<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=utils
-
agent 3.04function upper_case that makes s upper case.<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=core
-
agent 3.04function smallest that returns the smallest number in nums.<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=stats
-
agent 3.03in tools.py that returns the first item.<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=tools
-
agent 3.02.py that returns the first letter of s.<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=utils
-
agent 3.01function echo that repeats s n times.<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=tools
-
agent 3.01py that returns the average of a and b.<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=helpers
-
agent 3.0py. It returns the first letter of s.<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=words
-
agent 3.0a function how_many that counts the items.<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=listlib
-
agent 2.99.py that prints "Hello, Zoe!".<|user_end|><|assistant_start|><|tool_call_start|>write<|arg|>path=greet