'in other words' restatement cues
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. 6 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 words that introduce a restatement, especially 'words' in 'in other words' and the colon in 'i.e.:'. It is a signal that the next sentence will explain the previous one in simpler terms. Turning it up did not produce a clear theme in the sample 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.
-
ESS is a general topic of conversation everywhere and in every field:- UNITY OF GOD'S ESSENCE, Unity of His attributes, and actions. In other word, also the unity of prophets and their teachings, the unity of the LAW, GH
-
that can be hosted by the CLR, while at the same time can be accessed in a uniform manner by all languages that the .NET platform support. In other words, CLS is a subset of the full functionality of CTS.⏎- Common
-
or title). If you highlighted a title with the data, you would use this option to prevent the program from sorting your title into the list. (In other words, the first row stays in place, regardless of its first letter).⏎When you
-
everything. Definition of terms that create your total calorie burn:⏎- RMR=resting metabolic rate-the amount of calories you need to exist without any activity-ie:bedrest⏎- TEA-thermic effect of activity-this, obviously,
-
you like. If you do find the new bulb to not give off enough light replace the new one with a higher “wattage” bulb. In other words instead of using a 820 lumen bulb to replace a 6
-
tool can prove to be very handy in the conversion of Arabic to Roman numerals. The tool basically works on a very simple concept of format conversion. In other words you just have to fill in the value of Arabic numeral and it will show the
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 book and the magazine "R. M. R. 3" and "R. M. R. 4" (and their subas) in the field of art and design. The design of these items is also a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the piezoelectric transducers point to their applicability to wide range of dynamic pressure field measurements
-
5 section were instrumented with custom-made pressure transducers using polyvinylidene fluoride (PVDF) piezoelectric polymer film. These transducers provided in situ measurement of impulse pressures. These
-
Livermore National Laboratory researchers said their new tool enhances the way computer chips, LEDs and transistors are build.Commonly used piezo-electric speakers, such as those found in a cell phone, operate at low frequencies that
-
the Chilean salt flats, the Salar de Atacama — have been photographed from the International Space Station.⏎CALIPATRIA — There’s a lot to see at Simbol Materials’
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
- L5:1847 suppresses
- L4:1559 suppresses
- L5:1424 excites
- L1:506 suppresses
- L5:1802 weakly suppresses
- L4:1498 weakly excites
- L4:1609 weakly excites
- L3:1847 weakly excites
- L3:2584 weakly suppresses
- L0:261 weakly excites
- L3:2565 weakly excites
- L4:2990 weakly suppresses
- L3:2135 weakly excites
- L1:2179 weakly suppresses
- L5:1870 weakly suppresses
- L4:301 weakly suppresses
- L0:2468 weakly suppresses
- L5:1161 weakly suppresses
- L1:379 weakly excites
- L2:2357 weakly excites
- L4:2018 weakly excites
- L2:2726 weakly excites 'by' after 'characterized' or 'marked'
- L2:2108 weakly suppresses
- L5:642 weakly suppresses
Feeds later layers
- L9:1921 excites
- L9:1624 weakly excites
- L13:2292 weakly excites opening parenthesis for a gloss or abbreviation
- L13:1216 weakly suppresses
- L16:2400 weakly suppresses
- L10:603 weakly suppresses
- L10:1142 weakly excites
- L12:2117 weakly suppresses
- L13:895 weakly excites
- L11:991 weakly suppresses
- L12:3032 weakly excites
- L8:2816 weakly excites
- L13:176 weakly suppresses
- L20:463 weakly suppresses
- L7:43 weakly excites
- L8:1707 weakly excites
- L20:2274 weakly excites
- L14:2610 weakly excites
- L12:1241 weakly excites
- L14:1527 weakly suppresses
- L11:2569 weakly excites
- L26:1388 weakly suppresses
- L26:697 weakly excites
- L10:2933 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
- _endian
- 网首页
- occer
- oned
- mandated
- anytime
- quam
- 前所
- alyzer
- oner
Suppresses
- _',
- *",
- ?',
- _",
- $",
- *',
- ',{
- ervative
- ",(
- ?",