Neuron Explorer

Qwen3-0.6B

"that" pointing back to a noun

A neuron in layer 3 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the word "that" when it points back to something just mentioned, as in "for that product" or "at that time". Turning it up made the writing wander without a clear link to this pattern.

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

    On June 26, the Chinese observed "an extraordinary star" that was described as greyish-yellow, and was "as large as a pellet." On that same date, David Rittenhouse (Norriton, Pennsylvania, USA)

  2. #2 peak

    “tracing request” to restore links with family members across the border via that platform.⏎[F5] The following day State A’s forces commence arrests of people

  3. #3 peak

    , and the effect may be determined only through the analysis of process establishment data for that product.⏎Initial temperature is normally determined using a dial, electronic or other type of

  4. #4 peak

    , while on the other, individual teams and firms can track the development of open source software through tools created for that purpose. There is a multitude that is available on the market, many of which

  5. #5 peak

    on September 10, then received the name of Hurricane Humberto at the first report of the National Hurricane Center (NHC) the next day. At that time the hurricane was located about 310 mi (500 km

  6. #6 peak

    bars denote the size of the inflation change in each item relative to its historical norm, measured by the standard deviation of the seven-month change in the inflation series for that item since 2000. (We go back only to 2

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: "I need to know 10,000,000,000,000,000,000,000,000,000,000

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

    ia in human culture⏎In the last quarter of the 18th century, Professor Jacob Seydelmann of Dresden developed a process to extract and produce a more concentrated form for use

  2. #2 peak

    the university. The Female Law Practitioners Act had been passed in 1896, and on 10 May 1897 Ethel Benjamin gave effect to the new legislation when she was admitted as a barrister and

  3. #3 peak

    that it competed with the figural component for the viewer’s attention, and eventually landscape developed into an independent genre. The paintings of German artist Albrecht Altdorfer mark a significant step in this evolution.⏎The majority of Altdor

  4. #4 peak

    92 she passed the university junior scholarship examination.⏎In 1893 Ethel Benjamin enrolled for an LLB degree at the University of Otago. She

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. L2:2922 weakly suppresses
  2. L1:1318 weakly excites
  3. L2:2170 weakly suppresses
  4. L2:126 weakly excites
  5. L0:1966 weakly suppresses
  6. L2:1215 weakly excites
  7. L2:977 weakly excites
  8. L2:2692 weakly excites
  9. L0:3033 weakly excites
  10. L2:527 weakly excites
  11. L1:685 weakly excites
  12. L1:2801 weakly excites
  13. L0:2691 weakly suppresses
  14. L0:444 weakly suppresses
  15. L0:288 weakly excites
  16. L0:1841 weakly suppresses
  17. L1:1029 weakly excites
  18. L0:2283 weakly suppresses
  19. L2:1481 weakly excites
  20. L2:853 weakly suppresses
  21. L0:149 weakly suppresses
  22. L0:1312 weakly suppresses
  23. L2:872 weakly excites
  24. L1:894 weakly excites

Feeds later layers

  1. L4:468 excites "that" after "than" or "like"
  2. L11:199 excites
  3. L4:731 excites
  4. L4:270 excites
  5. L17:1816 excites
  6. L4:968 weakly excites
  7. L7:562 weakly excites
  8. L12:521 weakly excites
  9. L8:147 weakly excites
  10. L17:2011 weakly excites
  11. L4:1105 weakly excites
  12. L22:2014 weakly excites
  13. L9:1231 weakly excites
  14. L9:82 weakly excites
  15. L17:221 weakly excites
  16. L15:470 weakly excites the word 'this' after a preposition
  17. L5:9 weakly excites
  18. L4:962 weakly suppresses
  19. L20:2038 weakly excites
  20. L15:537 weakly excites middle digits of numbers
  21. L18:731 weakly excites
  22. L13:459 weakly excites
  23. L15:1957 weakly excites
  24. L4:1762 weakly excites "this" in time and context phrases

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.

1 values between -1.00 and -0.921 values between -0.68 and -0.601 values between -0.51 and -0.438 values between -0.43 and -0.3529 values between -0.35 and -0.27122 values between -0.27 and -0.19438 values between -0.19 and -0.11943 values between -0.11 and -0.021181 values between -0.02 and 0.06715 values between 0.06 and 0.14309 values between 0.14 and 0.2299 values between 0.22 and 0.3024 values between 0.30 and 0.396 values between 0.39 and 0.471 values between 0.47 and 0.552 values between 0.55 and 0.634 values between 0.63 and 0.713 values between 0.71 and 0.793 values between 0.79 and 0.883 values between 0.88 and 0.962 values between 0.96 and 1.042 values between 1.04 and 1.122 values between 1.12 and 1.201 values between 1.28 and 1.372 values between 1.37 and 1.452 values between 1.45 and 1.531 values between 1.61 and 1.691 values between 1.86 and 1.941 values between 2.02 and 2.101 values between 2.18 and 2.26offnegativepositiveoften

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. 他认为
  2. 他就
  3. \Eloquent
  4. ”的
  5. aisal
  6. 之后
  7. 他会
  8. 他也
  9. 也好
  10. 但是他

Suppresses

  1. лиз
  2. orb
  3. ibility
  4. viron
  5. erable
  6. stile
  7. 完整性
  8. =&
  9. ancell
  10. shore