the word tibia (shin bone)
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 5 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron fires on the word tibia, the shin bone, and on a few other bone names. When turned up, the model's writing wandered without any clear link to bones, so only the text examples support this.
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 activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
any known injury at that time. Examination reveals full motion of the right hip, knee, and ankle. He has tenderness to palpation over the anterior tibia with minimal swelling. No erythema
-
works was the Libellus de abbreviaturis Hebraeorum, tam Talmudicorum quam Masoritarum et aliorum rabbinorum (Paris, 1561), later exploited by Guy *Le
-
one of the major ligaments of the knee that is located in the middle of the knee and runs from the femur (thigh bone) to the tibia (shin bone). It prevents the tibia from sliding out in front of
-
20 years of age, and stabilizes when the bones are no longer growing.⏎Almost all OFD tumors occur in the long middle part of the tibia, with a few, rare
-
1% of all cancerous tumors that begin in the bone are adamantinomas.⏎These slow growing tumors most often occur in the middle part of the tibia, and in many cases, the fibula bone is also affected. Although uncommon
-
distal tibial shaft with no significant swelling. Radiographs are shown in Figure A and B. What is the most appropriate treatment?⏎MRI of the tibia⏎Aspiration of the tibia⏎Referral to child services⏎Long leg
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
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.
Here is something I have been thinking about lately: the 1000.000.000.000.000.000.000.000.000.000.000.0
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
to the Hittites around 1150 BC. They had healthy relations with the Kassites and Hurrians. ⏎After the Battle of Megiddo (1121 BC) Egypt pushed Mit
-
saying - Write a review⏎Respiratory system 52 Thoracotomy⏎Stem cell transplantation⏎Cardiovascular system 56 Coronary artery bypass graft⏎Cardiac valve surgery⏎Gastrointestinal system 29 Gastro
-
in agents to conduct the land sale. In the midst of rich farmland, Whitewright soon drew settlers from the small, nearby communities that the railroad had bypassed
-
by the team. Moreover, QA is about engineering processes that assure quality is achieved in an effective and efficient way.⏎What should be on a QA tester’s résumé?⏎For instance, if a defect is found and fixed, there is no
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L4:591 strongly excites
- L4:524 strongly suppresses
- L2:598 strongly suppresses
- L4:643 excites mid-word pieces of technical words
- L2:1853 excites
- L3:2273 excites
- L4:814 suppresses
- L4:345 suppresses
- L4:1345 suppresses
- L2:1091 excites
- L4:609 suppresses
- L4:1214 suppresses
- L0:1395 suppresses
- L4:1581 excites
- L2:761 excites
- L1:259 excites
- L4:1045 excites
- L2:1202 excites
- L2:801 excites
- L4:490 excites decade digit in years like 1930s
- L2:1969 excites
- L2:1253 suppresses
- L4:2239 suppresses
- L3:2921 excites
Feeds later layers
- L6:177 strongly suppresses
- L10:94 strongly excites
- L16:1231 strongly excites
- L6:853 excites
- L19:395 excites
- L22:1055 excites
- L21:625 excites
- L24:118 excites
- L21:327 excites
- L17:446 excites
- L25:1493 excites
- L15:2378 excites
- L18:1399 excites
- L16:286 excites
- L6:2548 excites
- L8:1546 excites
- L18:931 excites
- L8:423 excites
- L17:647 excites
- L22:271 suppresses
- L8:71 excites
- L15:2289 excites
- L6:1100 excites
- L19:1059 excites
Activation histogramA chart of how often a neuron takes each activation value.
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.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
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 layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- theless
- ly
- ️
- ic
- gether
- as
- ов
- ful
- ся
- empre
Suppresses
- Vish
- plotlib
- //-
- plies
- 砧
- )[-
- Prosper
- wegian
- tôn
- .mj