unclear: mid-name word fragments
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. 3 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 fires on the first or middle piece of a longer word or name, mostly in encyclopedic history text. The pattern is loose and the writing it produces under steering does not show a clear theme.
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.
-
, Erie, PA, was one of the few non-mountainous spots in<|endoftext|>The New Student's Reference Work/Alcinous⏎Alcinous (al-cĭn' o-us), a mythical king of the island of Scheria
-
into a dozen burial towers, the city's two major sanctuaries — the Temple of Bel (God) and the charming Temple of Baalshamin (god of fertility) — and a monumental arch that opens onto a 1,2
-
the Winyaw Sanitarium for the treatment of diseases of the lungs and throat. Modeled and conducted along the same lines as those in Germany, the sanitarium soon was receiving patients from all over the United States.⏎After Koch discovered
-
'just' a children's game.⏎"At Somerset's Punkie Night, at the end of October, children made punkies (lanterns) out of mangel-wurzels (a large kind of beet) and went knocking on
-
, who denounced King as a lawbreaker who had been manipulated by Communists. The situation looked bleak.⏎By then, Wonder had matured from a young harmonica-playing sensation to a chart-topping music genius lauded for his complex rhythms and
-
, Kalingrad Oblast, Russia), also called "the scattered church of Prussian Lithuania." The Mennonites living here had much to endure from the dragoons of the Prussian King Friedrich Wilhelm I, who finally expelled them from
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: GERMANIA, THE THEOROMACHY, 22STONIANOPOLE, FLUIDICS, THE ONLINE ### 2024, 22STONION A... A...
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ore of philadelphia who donated two chromebooks to kids at the gloucester As covid 19 cases surge across ontario hundreds of elementary students across the toronto district school board have been Children who attend preschool enter kindergarten with improved skills but
-
and traditions of their people⏎- they organized forces to win independence⏎what happened in the russo-japanese war and what was the outcome?⏎- japan crushed russia⏎- japan overran most of southeast asia during WWII, from indoch
-
the silver triumph of freedom…⏎Douglass marked down a gradual process of how his psyche driven to
-
to all of the courses in the specialization, and you earn a certificate when you complete the work.⏎for college students (except those in canada), log in to sapling learning, your online homework solution. in this course, you will learn about
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
- L1:71 strongly excites
- L2:3 strongly suppresses
- L1:77 strongly excites
- L2:495 strongly excites
- L2:201 excites
- L2:381 excites
- L0:138 suppresses
- L1:655 suppresses
- L0:212 excites
- L0:130 excites unclear: word right after a modifier
- L1:1154 suppresses
- L2:882 excites
- L2:59 excites
- L1:762 suppresses
- L1:893 suppresses
- L2:483 suppresses
- L1:824 suppresses
- L2:969 suppresses
- L2:2247 suppresses
- L1:122 excites 'World War II' and the 'II' after 'War'
- L0:115 suppresses
- L2:74 weakly excites
- L2:70 weakly excites
- L0:1301 weakly suppresses
Feeds later layers
- L4:176 strongly excites
- L4:56 strongly suppresses
- L10:482 suppresses
- L8:637 suppresses
- L5:429 suppresses
- L7:280 suppresses
- L9:2624 excites
- L4:158 suppresses
- L4:1601 suppresses
- L9:450 suppresses
- L6:2656 suppresses
- L5:437 suppresses
- L5:2508 suppresses
- L6:505 suppresses
- L4:1414 suppresses
- L4:662 excites
- L4:577 suppresses
- L6:2959 excites
- L4:0 suppresses
- L5:1286 excites
- L7:272 suppresses
- L6:2685 suppresses
- L4:1740 suppresses
- L20:987 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
- ronym
- uteur
- ующ
- itative
- ponsible
- unist
- reated
- ourced
- INCIDENTAL
- reating
Suppresses
- 痣
- tran
- 不舒服
- VILLE
- .port
- ford
- Isle
- lle
- XmlDocument
- 岛