'gl' starting words like glutes
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
It fires on the 'gl' sound at the start of a word split into pieces, such as 'glutes', 'glands', 'glaucoma', 'glia' and 'glint'. It turns down on words about prices and costs, like 'price' and 'cost'. Turning it up made the model repeat the word 'sound' or drift into unrelated lines.
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.
-
also strengthen your upper body.⏎Move More Muscles⏎Most of the muscles you use in running are in your lower body: your quads, hamstrings, glutes, hip flexors and calves
-
and lower-body muscles serve as your primary movers, and you strengthen many more muscles than when running. In addition to your calves, hamstrings, quadriceps and glutes, you strengthen your abs and erector spinae in your core,
-
-white discharge. There may be spread on to the perineum and into the groins. In Candida balanitis, tiny papules develop on the glans penis, evolve as white pustules or vesicles and rupture, leaving
-
wash your hands often, and avoid touching your eyes.⏎This is especially true if you have had any eye surgery such as LASIK, cataract surgery or glaucoma surgery, since your eyes may be more prone
-
85 percent.⏎The group name for white matter cells, glia—derived from the world ‘glue’—reflects their lowly
-
claimed to have been knocked out by a glancing blow, and said that when he came to, he found that the cook
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: sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound sound
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
is the actual rate of price inflation. They don’t use the CPI ratio put out by the Census Bureau. They use what is called a “chained” price deflator from the GDP report which means they take prices as they were in
-
deviation of the seven-month change in the inflation series for that item since 2000. (We go back only to 2000 because price indexes for some of the twenty-three items are not available before then.) Bars of
-
a reduction in volume per plant. Thus, total manufacturing costs have been increased about a cent a pound, compared with the same months a year ago.⏎Using typical cost figures supplied us from various types of creameries, it is evidence that it costs
-
market movement. It is one of the major components of the forward price index methodology. Variable cost index also known as VAT, is often confused with the expression “cost of good sale” which should not be confused as the same thing. The main
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
- L21:873 suppresses
- L3:1824 suppresses
- L9:1061 weakly excites
- L9:991 weakly suppresses
- L4:2237 weakly excites
- L10:1119 weakly excites
- L6:772 weakly excites
- L10:1042 weakly suppresses
- L10:1286 weakly suppresses
- L6:401 weakly excites
- L22:113 weakly excites
- L21:2455 weakly suppresses
- L16:126 weakly excites
- L15:1595 weakly excites
- L8:1596 weakly suppresses
- L21:1588 weakly excites
- L10:2195 weakly suppresses
- L24:151 weakly excites
- L7:2823 weakly excites
- L3:1346 weakly suppresses
- L19:171 weakly suppresses
- L16:2287 weakly suppresses
- L6:1978 weakly suppresses
- L6:2350 weakly suppresses
Feeds later layers
- L27:1613 weakly suppresses
- L27:772 weakly suppresses
- L27:2230 weakly excites
- L27:1331 weakly excites
- L27:1442 weakly suppresses
- L27:1657 weakly excites
- L27:1997 weakly suppresses
- L27:1904 weakly suppresses
- L27:2466 weakly suppresses
- L27:2510 weakly suppresses
- L27:2004 weakly excites the 's' after a hyphen
- L27:491 weakly suppresses
- L27:2214 weakly suppresses
- L27:1336 weakly suppresses
- L27:2999 weakly suppresses
- L27:1162 weakly excites
- L27:2878 weakly suppresses
- L27:1730 weakly suppresses
- L27:1688 weakly suppresses
- L27:991 weakly excites
- L27:369 weakly excites
- L27:2836 weakly excites
- L27:2416 weakly suppresses
- L27:2558 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
- imulation
- acial
- odata
- кра
- _SRV
- sábado
- ibrate
- クラ
- RVA
- bearer
Suppresses
- izados
- plat
- sound
- ertainment
- Comput
- ological
- chain
- ennis
- maxlen
- ��