'poly' word-start fragment
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. 24 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
Fires on the start of chemistry and materials words that begin with 'poly', like polyethylene, polymer and polyol. Turning it up made the model produce strings of digits and odd repeats, so the steering result does not clearly reflect this pattern.
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.
-
ubiquitylation system.⏎Ubiquitin can be ubiquitinated on seven lysine (Lys) residues or on the N-terminus, leading to polyubiquitin chains that can encompass complex topologies. Alternatively, or in addition
-
This is a question that many consumers have when it comes to choosing plastic bottles for their food and beverage products.⏎PET, or polyethylene terephthalate, is a type of plastic that is commonly used
-
. The resulting sensor array then was integrated into a microfluidic chip made of poly(dimethylsiloxane), a standard material used to make lab-on-a-ch
-
qPCR (Figure 1). Endogenous miRNAs are not polyadenylated, therefore during the reverse transcription process, a poly(A) polymerase polyadenylates the miRNA. The miRNA is used as a template to
-
RNA constitutes approximately 90%. It is challenging to enrich messenger RNA from a metagenomic microbial RNA sample because most prokaryotic mRNA species lack stable poly
-
carbyne (about 0.71 nm). Our approach of HV high temperature annealing has a great advantage over filling SWCNT/DWCNT by poly
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: once in a while, we get a moment to once again get at once again once again once again at once again once again once again once again once again at once again once again at once again once again once again once again once again once
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
use, and are used raw. The flowers contain a volatile oil made of a variety of amino acids and other substances, including borneol, camphor, adenine, and small amounts of vitamin B1.⏎In traditional Chinese medicine, ch
-
by using chemicals to kill germs on surfaces or objects. According to the CDC, several types of disinfectants are effective against the flu including chlorine bleach, hydrogen peroxide, and quaternary ammonium-based cleaners. No matter which product
-
and perirectum also are affected. Signs of infection, including warmth and swelling, may be absent.⏎Neutropenia also suggests the following possibilities:⏎Aplastic anemia reduces the bone marrow's ability to produce white blood cells.⏎Auto
-
:⏎|Vitamin B12 Need||94%||Confirm|⏎|Lupus (SLE)||2%||Ruled out|⏎|Aplastic Anemia||1%||Ruled out|⏎|Fanconi Anemia
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
- L22:819 suppresses
- L23:389 suppresses
- L20:2701 excites
- L1:180 excites
- L22:2552 weakly suppresses
- L23:2273 weakly suppresses
- L16:2589 weakly excites
- L0:356 weakly excites
- L21:1590 weakly excites
- L7:754 weakly excites
- L20:3034 weakly suppresses
- L1:363 weakly excites
- L2:1207 weakly excites
- L10:2671 weakly suppresses
- L5:942 weakly excites
- L11:2432 weakly excites
- L22:2483 weakly excites
- L7:287 weakly suppresses
- L20:2513 weakly excites
- L13:937 weakly suppresses
- L5:2695 weakly excites
- L8:1367 weakly excites
- L10:2954 weakly excites
- L5:135 weakly suppresses
Feeds later layers
- L25:1749 weakly excites
- L25:1700 weakly excites
- L25:759 weakly excites
- L25:1524 weakly excites
- L27:1247 weakly excites
- L27:991 weakly suppresses
- L27:2747 weakly suppresses
- L26:2062 weakly suppresses
- L25:605 weakly suppresses
- L25:1500 weakly suppresses
- L27:1252 weakly suppresses
- L25:1058 weakly excites
- L27:2846 weakly excites
- L26:2836 weakly suppresses
- L26:644 weakly suppresses
- L27:1571 weakly suppresses
- L26:2075 weakly suppresses
- L25:228 weakly suppresses
- L25:1354 weakly excites
- L25:2520 weakly suppresses
- L25:2440 weakly excites
- L27:887 weakly suppresses
- L27:485 weakly excites
- L27:519 weakly 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
- 解放
- ac
- utt
- ad
- AD
- aint
- und
- adol
- Submission
- .toHexString
Suppresses
- reap
- _SI
- iability
- LAG
- misc
- już
- Mis
- vero
- .mi
- lã