Feature #2774
Fires on “ result” (not labeled yet: read the examples)
- Fires on
- 0.47% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 90%
- story 9%
- python 1%
- chat 1%
- worksheet 0%
Makes the model say
What't heavy hard out differentCreated))
Makes it not say
friends friend came bunny's sun stopped (
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.83⏎9: for n in nums:⏎10: result = result + n
-
agent 2.7911: for n in nums:⏎12: result = result + n
-
agent 2.66⏎9: for n in nums:⏎10: result = result + n
-
agent 2.610⏎3: for n in nums:⏎4: result = result +
-
agent 2.611⏎3: for n in nums:⏎4: result = result + n
-
agent 2.6: acc = acc + i
-
agent 2.580⏎3: for n in nums:⏎4: result = result + n
-
agent 2.5110: for n in nums:⏎11: result = result +
-
agent 2.511⏎7: for n in nums:⏎8: result = result + n
-
agent 2.4910: for n in nums:⏎11: result = result +
-
agent 2.480⏎7: for n in nums:⏎8: result = result + n
-
agent 2.480⏎3: for n in nums:⏎4: result = result + n