Neuron Explorer

Qwen3-0.6B

'at' after a job title, before an institution

A neuron in layer 21 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 'at' when it comes right after someone's job title and before the place they work, such as 'a researcher at Stanford' or 'a professor at the University'. Turning it up made the model write 'University of Arizona' and similar university names over and over.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    morphine and binds to the same receptor, but the way it turns the receptor on is slightly different," says Dr. Aashish Manglik, a researcher at Stanford University School of Medicine who studies opioid receptors.⏎After Bohn's discovery,

  2. #2 peak

    said Guillaume Gronoff, a French astrophysicist at NASA Langley.⏎"And you think, 'OK, what I am looking at

  3. #3 peak

    twice as many goods to China each year.⏎"Both sides are yet to tap their trade potential," said Ravi Shekhar Vishal, an assistant professor at Sikkim University who co-authored a research paper on Nathula trade. "

  4. #4 peak

    kidney diseases, as well as provides information on a protein region whose interaction with filtered light chains is implicated in cast nephropathy”, says Professor Luca Jovine at the Department of Biosciences and Nutrition and corresponding author of the paper.⏎U

  5. #5 peak

    ’s largest anti-masking campaigns, spearheaded by the Anti-Mask League of San Francisco, according to Howard Markel, a professor of the history of medicine at the University of Michigan and co-editor-in chief of The American Influenza

  6. #6 peak

    ensure that resources are allocated to patients with the greatest need,” says Natalia Trayanova, the Murray B. Sachs Professor in the Department of Biomedical Engineering at The Johns Hopkins University Schools of Engineering and Medicine and the project’s principal investigator…

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: when you really, really, really, you just need to get here, you need to get here, you need to get here, you need to get here, and you need to get here. That is, when you really, really

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

    Salby told the H<|endoftext|>Posted by Marie on .⏎Pilot Reem Ahmad's flight plan has her leaving Dulles International Airport and flying a Boeing 727 due east at 450 mph. There is a 65

  2. #2 peak

    the Advanced Research Core (ARC) of the Concussion Assessment, Research, and Education (CARE) Consortium. The consortium is a joint effort of the National Collegiate Athletics Association (NCAA) and the U.S. Department of Defense.⏎

  3. #3 peak

    th ed.). Boston, MA: Cengage Learning.⏎[TradingCoachUK]. (2018, July 6

  4. #4 peak

    October 2016, we will conduct a series of low-speed, low-altitude flights across the Great Pacific Garbage Patch in a C-130 Hercules aircraft, using expert spotters and state-of-the-art sensors.

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. L19:2937 suppresses unclear: mixed possessives and fragments
  2. L18:1931 suppresses
  3. L19:2246 suppresses
  4. L18:3001 excites
  5. L17:3041 excites
  6. L17:978 excites
  7. L20:2623 suppresses
  8. L19:2135 suppresses
  9. L18:2529 excites
  10. L10:3047 excites second word of publication names
  11. L11:1994 excites
  12. L18:1451 excites
  13. L20:1661 excites
  14. L19:2255 suppresses
  15. L12:1853 excites
  16. L16:2427 suppresses
  17. L6:2759 excites
  18. L13:3005 suppresses
  19. L12:2889 suppresses
  20. L16:2852 suppresses 'in' after a noun or participle
  21. L18:2292 suppresses
  22. L13:894 excites 'of' after a person's name
  23. L20:2169 excites
  24. L17:1522 suppresses

Feeds later layers

  1. L22:2134 excites
  2. L22:838 excites
  3. L22:1083 excites
  4. L24:1210 excites
  5. L25:3053 excites
  6. L24:1424 excites
  7. L24:340 excites
  8. L25:875 excites
  9. L23:330 suppresses
  10. L24:2460 excites
  11. L22:2339 excites
  12. L24:1124 excites
  13. L23:348 excites
  14. L23:1501 excites 'at' in idioms like 'at work'
  15. L22:1981 weakly excites
  16. L27:2 weakly excites
  17. L26:404 weakly excites
  18. L24:2988 weakly excites
  19. L23:2719 weakly excites
  20. L22:1340 weakly suppresses
  21. L27:1886 weakly excites
  22. L26:23 weakly excites
  23. L25:2475 weakly excites
  24. L24:1953 weakly 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.

1 values between -8.31 and -6.805 values between -5.30 and -3.8013 values between -3.80 and -2.30222 values between -2.30 and -0.793475 values between -0.79 and 0.71109 values between 0.71 and 2.2129 values between 2.21 and 3.7120 values between 3.71 and 5.223 values between 5.22 and 6.725 values between 6.72 and 8.224 values between 8.22 and 9.725 values between 9.72 and 11.223 values between 11.22 and 12.731 values between 12.73 and 14.231 values between 14.23 and 15.731 values between 15.73 and 17.232 values between 18.73 and 20.241 values between 20.24 and 21.741 values between 23.24 and 24.742 values between 26.25 and 27.751 values between 32.25 and 33.761 values between 36.76 and 38.261 values between 38.26 and 39.761 values between 44.27 and 45.771 values between 50.28 and 51.78offnegativepositiveoften

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. 中信
  2. Univers
  3. University
  4. University
  5. ATIC
  6. SIN
  7. institutions
  8. Univers
  9. universities
  10. university

Suppresses

  1. notated
  2. ozo
  3. Zombies
  4. 老人家
  5. OPSIS
  6. -mode
  7. izador
  8. subclass
  9. bootstrap
  10. 节目中