Feature #3138
Fires on “music” (not labeled yet: read the examples)
- Fires on
- 0.61% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 75%
- python 13%
- story 8%
- chat 4%
- worksheet 0%
Makes the model say
redf elseomwhere I",
Makes it not say
? "result returns becameRead name rain
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.74py", line 3, in <module>⏎ assert combine("tiger", "tiger")
-
agent 3.73, echo, count_vowels⏎⏎ Line 4: assert echo("hello", 3)
-
agent 3.68import length, welcome, reverse⏎⏎ Line 3: assert length("apple") == 5
-
agent 3.651: from textlib import echo⏎⏎ Line 3: assert echo("cat", 3)
-
agent 3.58, repeat, first_letter⏎⏎ Line 4: assert repeat("banana", 2)
-
agent 3.56difference⏎3: ⏎4: print(join_words("music", "banana"))
-
agent 3.48py", line 3, in <module>⏎ assert combine("cloud", "moon")
-
agent 3.46words import initial, echo⏎⏎ Line 4: print(echo("music", 3))
-
agent 3.45from strings import exclaim, size⏎⏎ Line 5: assert size("river") == 5
-
agent 3.42import exclaim, whisper, size⏎⏎ Line 3: assert exclaim("cloud") == "cloud
-
agent 3.4difference⏎3: ⏎4: print(join_words("music", "banana"))
-
agent 3.39run.py<|arg|>oldString=print(joni_words("house", "moon"))