Neuron Explorer

Qwen3-0.6B

the word "thought" before "to"

A neuron in layer 2 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 "thought" in phrases like "is thought to", where a scientific claim is being hedged. The negative side fires on words like "met" or "high" that complete a condition. Turning it up did not give a clear change in the writing.

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

    or fire-adapted savanna-like habitats. These habitats are similar to the steppe and thorn scrub environments of western North America in which the genus is thought to have evolved.⏎Individual tortoises can live to be 60 years

  2. #2 peak

    electro-reduction studies date back to experiments done on zinc more than a century ago.2, 3 In general, large overpotentials are required that are thought to result from a reduced CO2–intermediate that is only partially stabilized by

  3. #3 peak

    to growth and development (Figure 1). At the heart of this clock is a feedback loop of gene expression known as the ‘central oscillator’, whose interaction is thought to regulate biological rhythms governing various physiological processes.⏎With their finding, the researchers have

  4. #4 peak

    may already be in serious or critical condition. The effect of these standard amounts of radiation certainly isn’t thought to be harmful, but there also haven’t been widespread studies related to their effects

  5. #5 peak

    anols — compounds like epicatechin, catechin and procyanidins, which are extracted from cacao beans and are thought to be responsible for the formation of nitric acid in the body. Nitric

  6. #6 peak

    CS began in 1998 and ended at the beginning of the study period. These conditions were thought to be of particular interest in assessing the influence of dredging activities on algal

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 am going to think about the 100, and I am going to think about the 100, and I am going to think about the 100, and I am going to think about the 1

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

    attributes. Each node in the tree represents a decision based on a feature, and the tree continues to split the data until a stopping criterion is met. The goal is to find the best split that maximizes the predictive accuracy of

  2. #2 peak

    International Renewable Energy Agency (IRENA), desalination with renewable energy can already compete cost-wise with conventional systems in remote regions where the cost of energy transmission is high. Elsewhere, it is still generally more expensive than desalination plants using

  3. #3 peak

    to carry on spreading the message about HIV.⏎Jaroslav Jedlicka, AIDS Programme Manager at Prague's National Institute of Public Health, says awareness is relatively high in the Czech Republic. However, he believes people are also becoming complacent:⏎

  4. #4 peak

    cells that are nearby detect these microRNAs, and when their concentration gets high enough, it causes those healthy cells<|endoftext|>Numerous women suffer with a range of

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. L1:433 suppresses
  2. L1:492 excites
  3. L1:1004 weakly suppresses
  4. L1:1492 weakly suppresses
  5. L0:1369 weakly suppresses
  6. L1:364 weakly excites
  7. L0:717 weakly suppresses
  8. L1:2340 weakly suppresses
  9. L1:386 weakly suppresses
  10. L1:2899 weakly suppresses
  11. L0:950 weakly excites
  12. L0:440 weakly suppresses
  13. L1:1790 weakly excites
  14. L1:541 weakly excites
  15. L1:532 weakly excites
  16. L1:1691 weakly suppresses
  17. L1:124 weakly suppresses
  18. L0:2157 weakly excites
  19. L0:1482 weakly excites
  20. L1:268 weakly excites
  21. L1:778 weakly suppresses
  22. L0:156 weakly excites
  23. L0:3030 weakly suppresses
  24. L1:759 weakly excites

Feeds later layers

  1. L3:4 excites
  2. L3:40 suppresses
  3. L4:160 weakly excites
  4. L3:534 weakly suppresses
  5. L3:471 weakly suppresses
  6. L8:408 weakly excites
  7. L3:27 weakly suppresses
  8. L24:991 weakly suppresses
  9. L6:93 weakly excites
  10. L10:984 weakly excites
  11. L11:1685 weakly excites
  12. L9:777 weakly excites
  13. L5:2360 weakly suppresses
  14. L26:234 weakly suppresses
  15. L5:1532 weakly suppresses
  16. L15:1313 weakly suppresses
  17. L6:418 weakly suppresses
  18. L25:1284 weakly excites
  19. L3:1283 weakly suppresses
  20. L24:200 weakly excites
  21. L10:1274 weakly excites
  22. L6:1332 weakly excites
  23. L11:3059 weakly excites
  24. L17:119 weakly suppresses

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 -1.11 and -1.051 values between -0.84 and -0.791 values between -0.79 and -0.731 values between -0.73 and -0.681 values between -0.68 and -0.632 values between -0.63 and -0.573 values between -0.57 and -0.521 values between -0.52 and -0.478 values between -0.47 and -0.4210 values between -0.42 and -0.3613 values between -0.36 and -0.3128 values between -0.31 and -0.2671 values between -0.26 and -0.20165 values between -0.20 and -0.15345 values between -0.15 and -0.10567 values between -0.10 and -0.041098 values between -0.04 and 0.01884 values between 0.01 and 0.06391 values between 0.06 and 0.12183 values between 0.12 and 0.1766 values between 0.17 and 0.2223 values between 0.22 and 0.2818 values between 0.28 and 0.339 values between 0.33 and 0.385 values between 0.38 and 0.444 values between 0.44 and 0.491 values between 0.49 and 0.543 values between 0.54 and 0.601 values between 0.65 and 0.701 values between 0.76 and 0.811 values between 0.81 and 0.861 values between 0.86 and 0.911 values between 0.97 and 1.02offnegativepositiveoften

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. отов
  2. otas
  3. esian
  4. ierge
  5. dings
  6. pher
  7. OTA
  8. ynth
  9. orph
  10. ::__

Suppresses

  1. 身后
  2. reamble
  3. ейств
  4. atorio
  5. körper
  6. ари
  7. 年后
  8. 特征
  9. 那一刻
  10. abo