dots in initials and abbreviations
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 very consistent: they mostly 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
The neuron fires on the full stop that follows a single letter, like the dot in a person's middle initial ('A. M. Sullivan') or in 'M. D. Anderson'. Turning it up did not make the writing about names or initials in any obvious way.
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.
-
0000000 SM: 11111111.11111110.00000000.00000000 IP: 1
-
Logic of Meaning and Meaningful Implication. B. Matalon(translated by J. Byrnes), A Genetic Study of Implication. R.J. Falmagne, Language and the Acquisition of Logical Knowledge. M.D.S.
-
on materials provided by NASA's Marshall Space Flight Center. Note: Materials may be edited for content and length.⏎Cite This Page:<|endoftext|>By A. M. Sullivan⏎CHAPTER IX. (continued)⏎From the Atlas and Cyclopedia of Ireland
-
d. Part V. Book IV. The Tissues and MechanNew edition in preparation. isms of Reproduction. Appendix. By A. S. LEA. 75. 6d. HANDBOOK OF
-
initial post(s) early in the module.⏎See the Discussion Board Guidelines and Rubrics⏎Download Discussion Board Guidelines and Rubrics.⏎Reh, F. J. (2018, November 01). Here Is a Look at
-
. Düchting). Computer simulation of the morphology and development of several species of seaweed using Lindenmayer systems (J.D. Corbit, D.J. Garbary). Generating fractals from Voronoi diagrams (K.W.
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 number of, the number of, the number of, the number of, the number of, the number of, the number of, the number of, the number of, the number of, the number of, the number of,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
comparing & contrasting modalities (a piece of lit to a video), or may do well in geometry…. Graphic organizers are a great resource for them⏎👩👩👦👦 Some enjoy working in
-
the full content.⏎This preview has intentionally blurred sections. Sign up to view the full version.View Full Document⏎Unformatted text preview: 60 | | | | CHAPTER 1 FUNCTIONS AND MODELS Not all functions possess inverses. Lets
-
even need to be real words, take, for example, Hanson's "MMMBop" or The Police's "De Do Do Do, De Da Da Da." Use whatever works.⏎Make compromises. You can take pre-existing lyrics and fit
-
.0/15 IP: 10101100.00010000.00000000.00000000 SM: 111
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
- L4:503 strongly suppresses
- L3:2985 strongly suppresses
- L4:330 strongly excites
- L3:1046 strongly excites
- L3:305 excites
- L1:2 suppresses
- L1:842 excites
- L0:7 excites
- L3:11 excites
- L2:2068 excites
- L2:878 suppresses
- L3:440 suppresses
- L0:14 excites
- L0:349 suppresses
- L0:508 suppresses
- L1:2115 suppresses
- L0:429 suppresses
- L0:617 excites
- L1:452 suppresses
- L4:235 weakly suppresses
- L2:47 weakly suppresses
- L0:732 weakly excites
- L0:188 weakly excites
- L0:1131 weakly excites
Feeds later layers
- L14:642 strongly suppresses
- L8:2572 strongly excites
- L9:1142 strongly suppresses
- L10:1091 strongly excites
- L13:1596 strongly excites
- L11:2666 strongly suppresses
- L11:1872 strongly suppresses
- L6:513 excites
- L11:1153 suppresses
- L6:1405 suppresses
- L17:1622 excites
- L15:344 suppresses
- L6:1839 suppresses
- L8:1435 excites
- L14:374 suppresses
- L9:146 suppresses
- L19:2663 suppresses
- L8:763 suppresses
- L12:1089 excites
- L10:85 suppresses
- L17:1590 suppresses
- L10:516 excites
- L13:1512 suppresses
- L12:1377 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
- pedia
- เหน
- ewood
- 之上
- 而来
- 之事
- 而成
- xa
- 但是如果
- 怙
Suppresses
- ادات
- /cms
- doen
- /do
- renc
- UDO
- angu
- weeds
- 朋友们
- udio