the letter f starting a word
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. 27 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 lights up on the letter f when it begins a word, like the f in fountains, flossing or firs. Turning it up mostly made the model's text fall into repeated loops rather than producing anything about f.
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.
-
through professional art design, you can draw a beautiful aquatic landscape.⏎In addition, different creative fountain forms are suitable for different application scenarios. For example, music creative fountains are generally suitable for meeting places such as squares. It is an organic combination
-
Older people who have used hearing aids will remember the days when the only way to change the settings on your hearing aid was to break out a screwdriver and fiddle around with the device. Even then, there were only a few different things
-
.⏎How to Take Care of Your Teeth with Braces⏎When you have braces, you'll have to be very good about following the brushing and flossing instructions your orthodontist gives you.⏎You can't eat hard,
-
offensive vapors are scrubbed or adsorbed before the exit gases are released into the hood exhaust system (see section 8.C.8.1 on fume hood scrubbers).⏎It is necessary to check if the hoods are performing
-
species than to others. In choice of nesting sites it exhibits a wide range of taste, and nests have been found in scrub oaks, pines, and firs, and varying in height from the ground from 3 or 4 feet
-
first time you ever hear of these things are when you see your hygienist. Did you also know that when your hygienist tells you to brush and floss more, it’s because this is the best way to prevent and treat these
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: Ifineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineineine
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
: territorial approach. Mechanical rules lead to certainty/predictability, but don’t take into account justice/fairness.⏎1. 10 states still follow Beale.⏎iii.
-
in cold-water fish.⏎- Steer clear of trans-fatty acids, which are found in deep-fried foods, margarine, and
-
Holy Spirit, who thus keeps the memory of Christ alive in his Church at prayer, also leads her toward the fullness of truth and inspires new formulations expressing the unfathomable mystery of Christ at work in his Church's life, sacraments,
-
Great Unknown: Risk-Taking Behavior in Adolescents⏎19.01.2017 | Max-Planck-Institut für Bildungsforschung⏎A sudden drop in outdoor temperature increases the risk of respiratory infections⏎
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
- L22:369 excites
- L18:437 excites
- L23:2441 excites
- L19:702 excites
- L2:1622 excites
- L18:328 suppresses
- L20:211 excites
- L0:755 suppresses
- L5:890 excites
- L18:569 excites
- L26:1564 excites
- L9:449 suppresses
- L17:378 weakly excites
- L24:586 weakly excites
- L20:1120 weakly excites
- L22:2428 weakly excites
- L5:91 weakly excites
- L21:2842 weakly excites
- L4:1892 weakly excites
- L16:2450 weakly suppresses
- L1:1464 weakly excites
- L21:237 weakly suppresses
- L1:951 weakly excites
- L23:206 weakly suppresses
Feeds later layers
None: this is the last layer.
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
- soever
- fs
- aked
- fed
- Hats
- flashes
- 一起
- .PNG
- bart
- fasta
Suppresses
- ine
- resh
- �
- безопас
- rax
- inely
- #pragma
- ulling
- Pub
- inding