unclear: weak, scattered activations
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. 4 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 barely responds to anything. Its strongest examples are weak and come from very different texts, so there is no clear pattern to describe. Turning it up gave mostly repetitive text.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
the Abyssins have not many images, they have great numbers of pictures, and perhaps pay them somewhat too high a degree of worship. The severity of their fasts is equal to that of the primitive church. In Lent they never eat till after
-
learner<|endoftext|>n.交響樂 [an elaborate instrumental composition in three or more movements, similar in form to a sonata but written for an orchestra (管弦樂隊) and usually of far grander proportions and more varied elements.]⏎
-
(formerly referred to as the reciprocal ohm or mho). Oliver Heaviside coined the term in September 1885.⏎Relation to other quantitiesEdit⏎As mentioned, conductance is related to resistance by:⏎for purely resistive
-
Journal suggests that it might be possible.⏎That's because a team of Swiss researchers found that when a hormone in the brain, erythropoietin (Epo), was elevated in mice, they were more motivated to exercise. In addition,
-
.What is the difference between MOV and MVI?⏎12. What are the functions of RIM, SIM, IN?⏎13. What is the immediate addressing mode?⏎14.
-
.<|endoftext|>Hamlet Essay, Research Paper⏎Hamlet is based on a young prince of Denmark that must uncover the truth about his fathers death. Hamlet a play that tells the story of a young prince who’s father recently died. Hamlets
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: "I am a number!" - and I will use the number to calculate the future value of the number. Now that is what? Let me know it. I have to calculate the future value of the number, but I am not.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
are carried over to the new data directory.⏎Stop the existing MySQL instance using the data directory. This must be a clean shutdown<|endoftext|>On April 16, 1853, the first train left Bombay for Thana. The
-
wrote in her diary, describing the tethered balloon being pulled along the National Mall. “He seemed to be enjoying his ride greatly.”⏎With this attention-grabbing stunt, Lowe was determined to
-
esite stratovolcano in the Islands of Four Mountains, a volcanic group in the Aleutian Arc. Like all stratovolcanoes, Mount Cleveland grew as explosive eru
-
really allowed to ask: “How many Americans have died after taking the COVID vaccine?”1⏎If you haven’t paid attention, the answer to this verboten (forbidden) question may shock you. Carlson points out (inaccurately
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
- L3:218 excites
- L3:604 suppresses
- L3:122 excites
- L3:464 excites
- L3:115 suppresses
- L3:188 excites
- L0:1301 suppresses
- L0:1224 suppresses
- L2:75 suppresses
- L2:111 suppresses
- L2:49 suppresses
- L3:59 excites
- L0:178 suppresses
- L2:1710 suppresses
- L3:456 suppresses
- L3:480 excites
- L3:1616 excites
- L3:2369 suppresses
- L3:294 excites
- L3:2543 excites
- L2:70 excites
- L3:50 excites
- L2:154 suppresses
- L3:421 suppresses
Feeds later layers
- L5:14 excites
- L5:230 suppresses
- L5:72 suppresses
- L5:2860 excites
- L5:81 excites
- L7:46 excites
- L15:1208 excites
- L6:49 excites
- L6:19 weakly suppresses second half of proper names
- L9:2567 weakly excites
- L5:69 weakly excites
- L16:194 weakly suppresses
- L5:414 weakly excites
- L16:454 weakly suppresses
- L15:402 weakly excites
- L5:129 weakly suppresses
- L6:2 weakly excites
- L9:3 weakly suppresses
- L14:270 weakly excites
- L14:2784 weakly suppresses
- L16:97 weakly suppresses
- L5:1052 weakly excites
- L12:994 weakly excites
- L6:413 weakly suppresses
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
- ny
- iy
- att
- alm
- 看起来
- ell
- oki
- sy
- SCORE
- ine
Suppresses
- .shtml
- ��
- erculosis
- icles
- writers
- aments
- urers
- itors
- wizards
- icians