Feature #1564
Fires on “ sum” (not labeled yet: read the examples)
- Fires on
- 0.37% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 45%
- agent 44%
- chat 9%
- story 2%
- worksheet 0%
Makes the model say
(appy rangeure len⏎ ⏎ astic
Makes it not say
<|assistant_end|> bright look her pretty this little b
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 4.72= return sum(nums) /
-
agent 4.11= return len(s)<|tool_call_end|>
-
python 3.9def biggest(nums):⏎ return max(nums)⏎
-
python 3.75(nums):⏎ return max(nums)⏎
-
python 3.75nums):⏎ return min(nums)⏎
-
python 3.74⏎ for i in range(n +
-
agent 3.62) / 2<|arg|>newString= return sum(nums) /
-
python 3.61_to(n):⏎ acc = 0⏎ for i in range(n +
-
python 3.59```python⏎def count_items(items):⏎ return len(items)⏎
-
agent 3.58newString= for i in range(1, n
-
agent 3.55item(nums):⏎ return max(nums)⏎
-
python 3.54`return sum(nums) /