Feature #17
Fires on “ mean” (not labeled yet: read the examples)
- Fires on
- 1.98% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 44%
- agent 30%
- story 15%
- chat 6%
- worksheet 5%
Makes the model say
_(!"return smiled return(" heard
Makes it not say
addsnder For foodNo]`.hingnums
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 2.9```python⏎def mean(a, b
-
python 2.73⏎def exclaim(s):⏎ return s + "!"⏎⏎⏎def add_one(x
-
python 2.69`result = result + 1`.⏎⏎```python⏎def add_all(nums
-
agent 2.67a, b):⏎ return a + b⏎⏎⏎def mean(a, b
-
python 2.667)` returns `-3`.⏎⏎```python⏎def only_even(nums
-
python 2.65to x.⏎⏎```python⏎def add_one(x
-
python 2.64x % 2 == 1`.⏎⏎```python⏎def even(x):⏎
-
python 2.626])` returns `6`.⏎⏎```python⏎def mean(a, b
-
python 2.61])` returns `4`.⏎⏎```python⏎def only_even(nums
-
python 2.54x % 3 == 0`.⏎⏎```python⏎def even(x):⏎
-
python 2.54s`.⏎⏎```python⏎def loud(s):⏎
-
python 2.54` with `return a ** b`.⏎⏎```python⏎def raise_to(a