Skip to content
mini-lab
Log in Sign up
Features of prelude-2

Feature #3482

Fires on “ many” (not labeled yet: read the examples)

Fires on
0.10% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 48%
  • worksheet 36%
  • agent 8%
  • story 7%
  • python 2%

Makes the model say

is + smallest thought tablely mean next

Makes it not say

_<|user_start|>ie"`.OneD$ For

Strongest examples

On text the dictionary never trained on. The underlined token is where it fires most.

  1. chat 6.42
    has 71 coins. Lily gets 8 more coins. How many coins does Lily have
  2. chat 6.24
    87 stars. Anna gets 872 more stars. How many stars does Anna have
  3. worksheet 6.16
    274 apples. Tom gets 39 more apples. How many apples does Tom have
  4. chat 6.1
    Sue has 3 cookies. Sue gets 4 more cookies. How many cookies does Sue have
  5. chat 6.09
    70 shells. Tom gets 47 more shells. How many shells does Tom have
  6. chat 6.08
    42 balls. Sue gets 18 more balls. How many balls does Sue have
  7. worksheet 6.0
    03 more stickers. How many stickers does Max have
  8. worksheet 5.95
    has 61 cookies. Max gets 1 more cookies. How many cookies does Max have
  9. chat 5.92
    63 coins. Sam gets 98 more coins. How many coins does Sam have
  10. chat 5.9
    Mia has 8 stickers. Mia gets 2 more stickers. How many stickers does Mia have
  11. chat 5.82
    1 more coins. How many coins does Lucy have
  12. chat 5.8
    Mia has 2 coins. Mia gets 6 more coins. How many coins does Mia have