Feature #2640
Fires on “ instead” (not labeled yet: read the examples)
- Fires on
- 0.25% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 78%
- python 16%
- story 5%
- chat 2%
- worksheet 0%
Makes the model say
home for]` place wind test can one
Makes it not say
vjsonoritch1onend
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.13in `arith.py`: it used `if a < b:` instead of `if a
-
agent 3.04number_utils.py`: it used `if a < b:` instead of `if a
-
agent 3.04py`: it used `return len(s) - 1` instead of `return len
-
agent 3.03py`: it used `return len(s) - 1` instead of `return len
-
agent 3.03in `calc.py`: it used `if a < b:` instead of `if a
-
agent 2.97in `greetings.py`: it used `return s[::1]` instead of `return s
-
agent 2.96in `calc.py`: it used `if a < b:` instead of `if a
-
agent 2.95.py`: it used `acc = acc + 1` instead of `acc
-
agent 2.94in `text.py`: it used `return s[::1]` instead of `return s
-
agent 2.92in `lib.py`: it used `if a < b:` instead of `if a
-
agent 2.91in `lib.py`: it used `return s[::1]` instead of `return s
-
agent 2.9_utils.py`: it used `result = result + i` instead of `result =