'a' in 'once a week' frequency
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. 7 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 small word "a" in phrases that give how often something happens, such as "once a week" or "twice a day." It is negative on phrases like "all at once" and "single session." Turning it up made the model drift into counting and number-heavy text, but not clearly about frequency.
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.
-
types of weeds, so add weeding the garden to your list of gardenia care tasks.⏎Part of caring for the gardenia is checking it at least once a week for signs of a white<|endoftext|>Disease induced changes in gene flow patterns among
-
bond price calculator:⏎Face value– The coupon's face value;⏎Coupon rate– The annual coupon rate;⏎Frequency– The coupon frequency, i.e., the number of times the coupon is distributed
-
.⏎“By the word of the Lord the heavens were made, and all the host<|endoftext|>Oxcarbazepine prevents fits.⏎Take it regularly, twice a day. Do not stop taking it unless your doctor tells you to.⏎It may
-
clean the smartphone regularly. This cleaning method can also be used with other devices such as laptop, and touched things. We should disinfect our devices at least once a day.⏎- Use your headphones⏎- Use your headphones instead of letting your smartphone
-
ystery' Bird Illness Continues, Tri-State May Put Bird Feeders Back Out With Precautions⏎The agency recommends thoroughly cleaning feeders at least once a week with a bleach solution and to take down your feeder for a week if you
-
infected at least once a day. Specific germ-prone areas such as bathrooms should also be given more attention
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 user's current understanding of the system, the potential for the system to handle more than what it's capable of. This is a thought of too much, too many ideas, too many things that are not just what the user can get
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
obstacles even if we don’t know it, providing our daily portion from His granaries in the mysterious world of the beyond. We are not given too much all at once
-
otaxy) is a type of minimally invasive surgical intervention. It uses radiation treatment to deliver a large and precise radiation dose to the tumor area in a single session. This type of surgery applies a three-dimensional coordinates system to locate small targets
-
if you try to do it all at once, you get overwhelmed," she said.⏎Don't get discouraged if you find
-
at all times. These cats’ stealth and adaptability allow them to survive well in a wide environment.⏎They’ve been known to jump 20 feet in one leap. The puma creeps up on its prey until it’s close enough
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
- L6:975 suppresses unclear: weak scattered triggers, finance on negative side
- L6:1638 excites
- L6:1963 suppresses
- L5:2095 weakly excites
- L2:639 weakly excites
- L4:2944 weakly excites
- L6:1453 weakly excites
- L5:1625 weakly suppresses
- L4:2440 weakly excites
- L6:1480 weakly excites
- L3:2814 weakly excites
- L5:2333 weakly excites
- L6:17 weakly suppresses
- L3:2050 weakly excites
- L3:1986 weakly excites
- L6:2421 weakly excites
- L5:311 weakly excites
- L6:12 weakly excites
- L0:1198 weakly excites
- L6:2700 weakly excites
- L2:153 weakly suppresses
- L5:2276 weakly suppresses
- L1:1163 weakly excites
- L6:2890 weakly suppresses
Feeds later layers
- L23:2496 excites
- L19:2715 excites
- L20:1155 excites
- L9:2069 weakly excites
- L24:1767 weakly excites
- L15:997 weakly excites
- L20:1335 weakly excites
- L9:1001 weakly excites
- L9:2327 weakly excites
- L24:1657 weakly suppresses
- L8:1923 weakly suppresses
- L25:1664 weakly excites
- L8:1850 weakly excites
- L22:1263 weakly excites
- L14:1164 weakly suppresses
- L22:2309 weakly excites
- L26:1055 weakly excites
- L8:1042 weakly excites
- L17:1656 weakly excites
- L10:2554 weakly suppresses
- L17:2356 weakly excites
- L20:981 weakly suppresses
- L17:1596 weakly excites
- L10:2536 weakly 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
- istrar
- inear
- periodic
- يق
- ometric
- quarterly
- unist
- ometers
- istributor
- ometer
Suppresses
- 罢
- material
- rag
- materials
- native
- material
- 抱
- eres
- enter
- 颚