word-start 'b' 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. 26 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
This neuron fires on the letter 'b' when it begins a word or a word piece, such as 'bakers', 'binging', 'bales' or 'bunnies'. Turning it down made the model get stuck repeating phrases, and turning it up gave off-topic text.
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.
-
On All Devices Downloads Are Available Free Of Charge As Long As You Do Not Remove The Link To English Practice At The Bottom Of Each Page Worksheets Are Divided Into B1 Intermediate B1 And B<|endoftext|>The Plain
-
the village to life with imaginative play. They can pretend to be doctors, musicians, firefighters, mail carriers, train engineers, construction workers, florists, and bakers. Children can also enjoy caring for babies
-
food. These obsessions and anxieties over food will negatively affect the person's mental, physical, and social health as they participate in activities like extreme diets, binging, purging, or avoiding certain foods.⏎An eating disorder is a behavioral
-
and an engine to power the equipment.⏎The stable to right of the gate housed draught horses that pulled drays of green flax from the swamp or delivered bales of fibre to Waikanae railway station.⏎Click the thumbnail to enlarge
-
Try peanut butter on apple slices or celery. Also try nuts, cheese and crackers, and low-fat dairy products like milk, cottage cheese, and yogurt.⏎- Bland foods, such as gelatin, frozen desserts,
-
! There’s more…⏎How Science Works has been adapted as an online course for middle- and high-school science educators to “broaden their own knowledge and understanding and to use with students, the course we
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 United States and China are both seen as both poor and poor-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at-at
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
workmanship they are made to last during His, not one another's pleasure.⏎This idea found its way into our Declaration of Independence in modified form:⏎We hold these truths to be self-evident, that all men are created equal, that
-
the two most important genealogies are recorded at Matt. 1:1–17 and Luke 3:23–28. These genealogies in the Gospels are essential, for they reveal the family lineage of Christ
-
as lower back (lumbosacral portion) or lower extremities.⏎Causes of chronic pelvic pain⏎Chronic pelvic pain may originate due to multiple etiological factors. The commonly studied causes of chronic pelvic pain are as follows-⏎-
-
. We must challenge every departure from them. There are just two tests: "Is this departure necessary to win the war?" "How are we going to restore these freedoms after the war?"⏎And the exploration of these questions calls for a calm and
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
- L22:939 strongly suppresses
- L22:809 suppresses
- L20:3051 excites
- L18:1821 excites
- L16:1330 excites
- L23:883 suppresses
- L21:338 suppresses
- L21:347 suppresses
- L18:1433 excites
- L19:2522 excites
- L17:963 excites
- L18:1556 suppresses
- L20:2948 excites
- L20:807 excites
- L16:1176 suppresses 'is' right after 'there' or 'nor'
- L23:119 excites
- L22:211 excites
- L17:2744 suppresses 'of' after 'those' or 'these'
- L19:1907 excites
- L19:1427 suppresses
- L20:2997 suppresses
- L19:2880 excites
- L19:1331 suppresses
- L6:156 suppresses
Feeds later layers
- L27:478 excites
- L27:208 excites
- L27:638 excites
- L27:1199 excites
- L27:814 excites
- L27:592 weakly excites
- L27:488 weakly excites
- L27:249 weakly suppresses
- L27:883 weakly excites
- L27:1177 weakly excites
- L27:758 weakly excites
- L27:645 weakly excites
- L27:2592 weakly excites
- L27:122 weakly excites
- L27:2629 weakly suppresses
- L27:533 weakly excites
- L27:503 weakly excites
- L27:2387 weakly suppresses lowercase m opening a word fragment
- L27:297 weakly suppresses
- L27:2107 weakly suppresses
- L27:1092 weakly excites
- L27:959 weakly excites
- L27:963 weakly excites
- L27:2786 weakly suppresses lowercase 'm' inside split words
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
- ivism
- arnation
- thood
- alysis
- notation
- discovery
- deviation
- nage
- usage
- achievement
Suppresses
- 雇
- laus
- _HC
- 用工
- ляем
- edic
- ималь
- (assigns
- \"",⏎
- gross