unclear: mid-word pieces inside names
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. 6 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 is weak and hard to read. Its highest reactions fall on pieces of longer names or technical words, like the middle of a species name or a place name split into parts. Turning it up or down didn't produce 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.
-
of the family Veneridae Rafinesque, 1815. Three species, including one new, are here described: Goshoraia minor Tashiro and Kozai, 1989 (Apt
-
in this way."⏎In their paper, published in the current issue of the Public Library of Science ONE, they say there's an established link between Type 1 diabetes and the body's ability to
-
ship shape = girlfriend friendship shipshape.⏎waist ? tail ? ? side ? ? fall ? ? down ? ? spring ? ? ? hole⏎Answerwaist coat tail gate way side arm pit fall out put down cast off spring time line man
-
referred to as Wolbachia is Wolbachia pipientis. Before DNA sequencing, when bacteria were classified morphologically, W. persica and W. melaphagi were also thought to be related, but these have nothing to do with
-
as:⏎- Rickets caused by vitamin D deficiency⏎- Pellagra caused by niacin deficiency⏎- Scurvy caused by vitamin C deficiency⏎- Beriberi caused by vitamin B1 (thiamine) deficiency⏎In the advent
-
. Latvia's neighbours to the east are the countries Russia and Belarus. Latvia is split into four parts called Kurzeme, Vidzeme, Zemgale, and Latgale.⏎History[change | change source]⏎Latvia was
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 have been thinking about lately, I have been thinking about lately, I have been thinking about lately, I have been thinking about lately, I have been thinking about lately, I have been thinking about lately, I have been thinking about lately
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
uses the resulting Internet address to connect to the web server, send the e-mail or use the World Wide Web.⏎When you visit http://www.internetsociety.org in a browser,
-
in space in detail, but they do indirectly contemplate the application of the law of armed conflict. Both the constitutive legal instruments and the subsequent efforts to address the current challenges to space security influence the way in which the law of
-
the process of amputating or cutting off a portion or all of the last bone of each digit/toe of the paw while the cat is under anesthesia.1-3 This is usually done at the pet owner’s request to prevent cats from
-
SYNDROME X DO? What does Metabolic Syndrome X cause? What are the effects of Metabolic Syndrome X? Over time, normally years, the metabolic imbalances give rise to a greater risk of cardiovascular disease, heart disease, kidney failure
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
- L4:278 strongly suppresses
- L2:39 strongly excites
- L2:69 strongly excites
- L4:129 strongly suppresses
- L1:140 strongly suppresses
- L4:695 excites
- L3:54 excites
- L3:280 excites
- L2:60 excites
- L0:144 suppresses
- L5:1341 suppresses
- L1:981 suppresses
- L4:436 excites
- L2:143 suppresses
- L2:33 excites
- L1:647 suppresses
- L1:434 suppresses
- L1:80 suppresses
- L2:195 excites
- L3:593 excites
- L5:29 suppresses
- L1:698 suppresses
- L0:96 suppresses
- L0:309 suppresses
Feeds later layers
- L9:1574 strongly excites
- L11:1243 strongly excites
- L7:461 strongly excites
- L12:831 strongly excites
- L10:1509 strongly excites
- L14:74 strongly excites
- L15:1550 strongly excites
- L8:154 strongly suppresses
- L9:421 strongly suppresses
- L12:1217 strongly excites
- L16:841 suppresses
- L8:654 excites
- L15:1583 suppresses
- L18:688 excites
- L12:396 suppresses
- L15:1058 suppresses
- L16:409 excites
- L8:630 suppresses
- L14:210 excites
- L16:654 excites
- L12:619 suppresses
- L16:766 suppresses
- L8:124 excites
- L9:342 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
- ros
- ROT
- syn
- _SOL
- ROL
- sol
- rays
- uada
- ALAR
- AMPL
Suppresses
- iki
- .dispose
- iver
- jak
- .escape
- hiểm
- aşağıd
- cube
- adjud
- —we