Feature #1551
Fires on “ head” (not labeled yet: read the examples)
- Fires on
- 1.25% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 60%
- agent 33%
- story 4%
- chat 2%
- worksheet 1%
Makes the model say
(!"ise theirto slDon because
Makes it not say
OneFix about minusred Try two arrived
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 6.36def head(items):⏎
-
python 6.13def tail(items):⏎ return items[-1]⏎⏎⏎def smallest(nums):⏎
-
agent 5.83% b == 0⏎⏎⏎def vowels(s):⏎
-
python 5.78def head(items):⏎ return items[0]⏎⏎⏎def biggest(nums):⏎
-
python 5.74x):⏎ return x % 2 == 1⏎⏎⏎def times(a, b
-
agent 5.64def tail<|tool_call_end|><|assistant_end|><|tool_start|>Found
-
python 5.61def half(x):⏎ return x // 2⏎⏎⏎def negate(x):⏎
-
python 5.57def half(x):⏎ return x // 2⏎⏎⏎def odd(x):⏎
-
python 5.57def initial(s):⏎ return s[0]⏎⏎⏎def echo(s, n
-
python 5.54⏎⏎⏎def square(x):⏎
-
agent 5.52newString= return b⏎⏎⏎def positive(x):⏎
-
python 5.521⏎⏎⏎def even(x):⏎