"of" after is/are/be (of + quality)
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. 17 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 word "of" when it comes right after a linking verb like "is", "are" or "be", as in "are of considerable importance" or "of two types". When it was turned up, the model's writing did not become clearly about anything in particular, so the steering results do not show a clear effect.
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.
-
) by rotating a magnetic field at a constant angular velocity within a stationary coil as shown in Fig. (a).⏎In either case, the generated voltage will be of sinusoidal waveform. The magnitude of generated voltage will
-
axes is perpendicular⏎to other axes and is of different length, e.g.⏎(iv) Orthorhombic System :⏎Three
-
is available for this purpose (e.g., US Census data at https://www.census.gov/) and can be of tremendous help in targeting various ethnic groups and geographic areas with specific fruit crops. Additionally
-
blackens the leaves; pull up the plants and discard them.<|endoftext|>Introduction and general survey⏎The economic importance of power amplifiers⏎Audio power amplifiers are of considerable economic importance. They are built in their hundreds of thousands every year, and
-
and should be of sufficient length to penetrate the muscle tissue. The anterolateral thigh can<|endoftext|>
-
interests of children in various age groups in order to write games that really appeal to the audience. In children's literature, for example, the protagonists tend to be of a similar age to their intended audience, a theme that game studios can apply to
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: I have a 2011 Mac 6000 (the first time I have a 6000) with a 60000-based 6000 (the first time I have a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
blood sugar. This can put you more at risk for Type 2 diabetes. This is one more reason that avoiding food all day and eating all of your calories near bedtime is not a good idea.⏎Another pattern that exists for people who eat late
-
and run without help. Enjoy every minute.⏎- FACT: 95 percent of 6- to 21-year-old students with disabilities attended public schools in 2007.⏎- TIP: If the baby does not crawl
-
TMJD as well.⏎What Causes Pain in the Temporomandibular Joints?⏎The temporomandibular joints (TMJs) are located in each ear, between the jawbone and skull. When you chew, your mouth
-
oceans. With over 80% of ocean debris ending in the ocean water, it impacts marine organisms. If not taken care of, plastics
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
- L16:2035 excites 'of' after a noun phrase
- L15:2682 excites
- L16:2291 suppresses
- L15:1223 weakly suppresses
- L6:2310 weakly suppresses
- L7:2705 weakly suppresses
- L12:2106 weakly suppresses
- L14:1492 weakly excites
- L15:1683 weakly suppresses
- L11:2840 weakly excites 'in' after a size measurement
- L15:2180 weakly suppresses
- L10:132 weakly excites
- L6:2156 weakly suppresses
- L6:1343 weakly suppresses
- L5:2462 weakly excites
- L5:47 weakly excites
- L16:2307 weakly excites
- L16:2564 weakly suppresses
- L16:1557 weakly excites 'in' meaning 'in this case' or 'in between'
- L5:1116 weakly suppresses
- L2:1323 weakly suppresses unclear: fragments of rare nouns
- L0:582 weakly suppresses
- L4:862 weakly suppresses 'to' after 'too big' or 'right angles'
- L2:851 weakly suppresses
Feeds later layers
- L18:1468 excites
- L18:1770 excites
- L19:1610 excites
- L22:1073 weakly excites
- L20:1494 weakly excites
- L21:444 weakly excites
- L19:964 weakly excites
- L19:248 weakly suppresses
- L19:769 weakly excites
- L18:1007 weakly excites
- L25:339 weakly excites
- L19:178 weakly excites
- L24:2680 weakly suppresses
- L18:1046 weakly excites
- L19:836 weakly excites
- L18:2885 weakly suppresses
- L19:1843 weakly suppresses
- L18:1690 weakly excites
- L21:662 weakly excites
- L19:2064 weakly excites
- L22:1689 weakly excites
- L18:1271 weakly excites
- L21:171 weakly excites
- L20:161 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
- entic
- verty
- чик
- awai
- atever
- alf
- enty
- posal
- inox
- etry
Suppresses
- 妞
- Die
- well
- bien
- アウト
- Phantom
- ('@
- 出局
- 敢
- 胼