Neuron Explorer

Qwen3-0.6B

cancer and tumor words in medical news

A neuron in layer 25 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 reacts to words about cancer and tumors, mostly in health news articles. When it was turned up, the model started writing the word 'biological' over and over, which is close to the medical and science theme.

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

    what we’re detecting with them is clinically meaningful,” said Lynn Sorbara, Ph.D., of NCI’s Division of Cancer Prevention.⏎Different Tests for Different Tumor Molecules⏎More than 100 years ago, scientists discovered that tumors

  2. #2 peak

    process that is easily incorporated in patient care," Venneti says. The study is published in Science Translational Medicine.⏎Traditional prognostic markers based on tumor grade have proven unreliable for ependymomas. Understanding a patient's prognosis can

  3. #3 peak

    often found mutated in non-smokers (EGFR). Led by HMS graduate student Stephen Curtis, the team identified cancer stem cells in a model combining the K-RAS and p53 oncogenic

  4. #4 peak

    .⏎With the array in hand, the investigators then attached a monoclonal antibody to the gold lining inside the holes. This monoclonal antibody binds to a cancer biomarker protein known as PAX8. The researchers then took advantage of an

  5. #5 peak

    Research, said in a briefing paper to the media this week.⏎M.D. Anderson researchers see leukemia and lymphoma blood-based cancers as the most amenable to cancer vaccines. But work also is ongoing to harness the immune system’s power to fight

  6. #6 peak

    the ongoing debate about benefits of primary cancer detection with cancer screening programs such as Mammograms, PSA screening, bowel cancer screening).⏎C

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: how to take a human image and show it in a more human image. How to show human images, human images, human images, human images, human images, human images, human images, human images, human images, human images,

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

    or acquired through infection.”⏎Finding a Heart Problem Before Birth⏎A congenital heart problem is a defect or disease that your child is born with. The most common are ventricular septal defects or atrial septal defects, which keep your child

  2. #2 peak

    well as an increased risk of atrial thrombus formation that may cause cerebral and/or systemic embolization.⏎Atrial flutter occurs in many of the same situations as atrial fibrillation, which is much more common. Atrial flutter may be

  3. #3 peak

    the basis of clinical suspicion or risky exposure. A human case of brucellosis was considered confirmed if results of serologic tests, such as ELISA or complement fixation test, were positive, in accordance with the World Health Organization case definition

  4. #4 peak

    -chain, medium-chain, or⏎long-chain acyl-CoA dehydrogenases; (2)⏎multiple acyl-CoA dehydrogenase deficiency (glutaric acid

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. L21:552 suppresses
  2. L21:205 suppresses
  3. L3:37 suppresses
  4. L20:958 excites
  5. L10:378 suppresses
  6. L20:150 excites
  7. L3:1091 excites
  8. L3:1037 excites
  9. L9:228 suppresses
  10. L8:39 suppresses
  11. L22:2115 suppresses
  12. L9:2044 weakly excites
  13. L3:768 weakly suppresses
  14. L14:717 weakly suppresses
  15. L16:2622 weakly suppresses
  16. L3:2581 weakly excites
  17. L9:2575 weakly suppresses
  18. L10:1768 weakly excites
  19. L4:1063 weakly suppresses
  20. L6:2234 weakly excites
  21. L20:2695 weakly excites
  22. L0:1955 weakly suppresses
  23. L6:1384 weakly suppresses
  24. L5:2326 weakly suppresses

Feeds later layers

  1. L27:1657 weakly excites
  2. L26:398 weakly suppresses
  3. L26:821 weakly excites
  4. L27:1572 weakly excites
  5. L27:1643 weakly excites
  6. L27:625 weakly excites
  7. L26:2716 weakly suppresses
  8. L26:1770 weakly excites
  9. L27:1322 weakly excites
  10. L27:2827 weakly suppresses
  11. L26:192 weakly excites
  12. L27:2762 weakly suppresses
  13. L26:2498 weakly excites
  14. L27:2542 weakly suppresses
  15. L27:2918 weakly excites
  16. L26:2916 weakly suppresses
  17. L27:903 weakly excites
  18. L27:2705 weakly excites
  19. L26:1694 weakly excites
  20. L27:444 weakly excites
  21. L26:2615 weakly excites
  22. L26:1696 weakly suppresses
  23. L27:1393 weakly suppresses predicting the word two
  24. L26:455 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.

4 values between -14.80 and -13.371 values between -13.37 and -11.943 values between -11.94 and -10.514 values between -10.51 and -9.087 values between -9.08 and -7.657 values between -7.65 and -6.2314 values between -6.23 and -4.8031 values between -4.80 and -3.3756 values between -3.37 and -1.94282 values between -1.94 and -0.513079 values between -0.51 and 0.92195 values between 0.92 and 2.3571 values between 2.35 and 3.7734 values between 3.77 and 5.2032 values between 5.20 and 6.6315 values between 6.63 and 8.0612 values between 8.06 and 9.4911 values between 9.49 and 10.926 values between 10.92 and 12.357 values between 12.35 and 13.774 values between 13.77 and 15.208 values between 15.20 and 16.636 values between 16.63 and 18.063 values between 18.06 and 19.493 values between 19.49 and 20.923 values between 20.92 and 22.352 values between 22.35 and 23.772 values between 23.77 and 25.202 values between 26.63 and 28.061 values between 28.06 and 29.491 values between 30.92 and 32.342 values between 40.92 and 42.34offnegativepositiveoften

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. biological
  2. Biological
  3. 生物
  4. 生物学
  5. biology
  6. bi
  7. biom
  8. Bi
  9. Biol
  10. Biom

Suppresses

  1. 圍
  2. erner
  3. atics
  4. оль
  5. 围
  6. 忍
  7. =obj
  8. 당
  9. udos
  10. venta