Neuron Explorer

Qwen3-0.6B

'with' or 'from' after comparison words

A neuron in layer 12 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the word that follows a comparison or agreement, such as 'compared with', 'compared to those', 'agree with' or 'a study from'. When it was turned up, the model's writing went off topic, so the steering does not clearly confirm it.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    might advocate for a client to consult with his or her psychiatrist.⏎Provide a specific example.⏎Be sure to support your postings and responses with specific references to the Learning Resources.⏎Lichtblau

  2. #2 peak

    less than 20% of its energy from ASFs compared to wealthier nations where ASF intake comprises about 40% of energy consumption.5 Literature from countries such as Kenya, where many of Australia’s Sudanese refugees lived in refugee

  3. #3 peak

    to balance our emotions… with studies also backing that a diet high in vegetables, fruit, unprocessed grains and fish is proven to lower levels of depression compared to those who eat a ‘Western’ processed diet. So, in a nutshell, what

  4. #4 peak

    Not-On-Tobacco (N-O-T) program in 8 Maryland high schools. The study was designed to compare N-O-T program stop smoking rates to those generated by control groups in which students were simply handed standard quitting brochures and

  5. #5 peak

    participants were tested on a standardized one-minute word reading test five times from the beginning of Grade 1 to the end of Grade 2. Results: As compared with the local norm, both Tier 1 and Tier students showed significant improvement after

  6. #6 peak

    heart disease and stroke. Grapefruit also contains natural vitamin C and potassium. A recent study from the Hebrew University's Hadassah Medical School in Israel performed both in vitro and

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

turned down a lot
Prompt

Here is something I have been thinking about lately: for the purpose of checking the accuracy of the results, I will calculate the mean of the results from a few different angles. Let's check the mean of the results, I will do the following: 1. Take the mean of the results

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    ium concentrations found in a buddingtonite sample had an RSD of 2.2% and their mean agreed with that obtained by the Kjeldahl method. By use of the prescribed dilution of the sample, detection limits of

  2. #2 peak

    determine all voltage drops using Ohm's Law (E=IZ ; I=E/Z ; Z=E/I). The polar magnitudes of the results should closely agree with your voltmeter readings.⏎Schematic with SPICE node numbers:⏎The two

  3. #3 peak

    angular and wavelength dependence of the reflectivity of uncoated, single-layer coated, triple-layer coated and graded-index-layer coated samples was performed and compared with experimentally derived films. Good agreement between model and experiment was observed.⏎ASJC Scopus

  4. #4 peak

    the viability of air injection as a means of mitigating cavitation erosion. The results from the impulse pressure measurements were compared with those obtained from hydrophone and accelerometer measurements. All the three measurement techniques showed similar results. The high sensitivity, minimal thickness

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L11:3051 suppresses
  2. L7:1095 excites
  3. L7:1516 suppresses
  4. L10:3044 suppresses
  5. L9:1990 weakly excites
  6. L10:2940 weakly suppresses
  7. L9:1805 weakly suppresses
  8. L8:1669 weakly excites
  9. L11:2411 weakly excites
  10. L9:2113 weakly suppresses
  11. L6:2518 weakly excites
  12. L10:1697 weakly suppresses
  13. L11:1918 weakly suppresses
  14. L9:2177 weakly excites
  15. L7:1924 weakly suppresses
  16. L1:2977 weakly excites
  17. L7:1326 weakly suppresses
  18. L9:1677 weakly excites
  19. L5:1795 weakly excites
  20. L0:2165 weakly excites
  21. L5:1889 weakly suppresses
  22. L10:2560 weakly suppresses
  23. L6:573 weakly excites
  24. L10:2670 weakly excites

Feeds later layers

  1. L14:2821 suppresses
  2. L14:2879 suppresses
  3. L16:1640 suppresses
  4. L13:1819 suppresses
  5. L14:2963 suppresses
  6. L15:3018 suppresses
  7. L13:2075 suppresses
  8. L14:2500 suppresses
  9. L13:1768 suppresses
  10. L14:1773 suppresses
  11. L13:2546 suppresses
  12. L14:2977 suppresses
  13. L13:2893 suppresses
  14. L17:1556 suppresses 'instead of' and 'rather than'
  15. L14:2428 suppresses
  16. L14:2802 suppresses
  17. L16:2197 suppresses
  18. L14:2976 suppresses
  19. L14:2429 suppresses
  20. L15:2309 suppresses
  21. L13:1728 suppresses
  22. L21:2248 suppresses
  23. L13:1058 suppresses
  24. L13:245 excites

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

2 values between -3.19 and -3.081 values between -2.19 and -2.081 values between -1.96 and -1.852 values between -1.41 and -1.304 values between -0.63 and -0.5123 values between -0.51 and -0.40109 values between -0.40 and -0.29357 values between -0.29 and -0.18855 values between -0.18 and -0.071368 values between -0.07 and 0.04751 values between 0.04 and 0.16330 values between 0.16 and 0.2767 values between 0.27 and 0.3830 values between 0.38 and 0.495 values between 0.49 and 0.601 values between 0.60 and 0.711 values between 1.05 and 1.161 values between 1.16 and 1.27offnegativepositiveoften

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.

Promotes

  1. ilitating
  2. <Select
  3. ngoing
  4. pher
  5. antor
  6. 会选择
  7. loys
  8. setSelected
  9. CharacterSet
  10. 男人

Suppresses

  1. 同类
  2. Ä
  3. error
  4. .Match
  5. coincide
  6. ổ
  7. discrepan
  8. errors
  9. contrad
  10. cliff