'be' after 'might', 'also' or 'only'
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. 1 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 word 'be' when it comes after words like 'might', 'also', 'only' or 'can't', as in 'you might also be interested'. Steering did not show a clear, consistent change in the 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.
-
cover creating a 90 degree pipe that goes with the T-Section pipe we created in the Creating A Simple Mesh Using Splines tutorial.⏎I will only be introducing one
-
⏎, focuses on Earth System science, including articles on student inquiry, differentiated instruction, geomorphic concepts, the rock cycle, and much more!⏎You might also be interested in:⏎About 70<|endoftext|>- freely available⏎Int. J.
-
others. Arid, desert climates have much less humidity and homes may need to be supplemented with a humidifier to make the air moister.⏎You might actually be the one responsible for frequent nose bleeds. Bad habits such as nose-picking
-
adequate resources: If a school doesn’t have enough funds to manage computers equal to number of students, trust me the idea will fail badly. Obviously you can’t be wasting the precious time of students getting them in turns to use the computers and requiring
-
disease,'' said Jack, echoing the opinion of the other 15 boys at the experimental Aspie school here in the Catskills. ''So we can't be 'cured.' This is just the way we are.''⏎From behind his
-
ots from forming.⏎Anti-platelet drugs are given to someone who has had a heart attack, to reduce the chances of them having another. You might also be given anti-platelets if you have had a stent fitted to a narrowed
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 can't be too much. I have to be a very small part of the whole. So, I need to be able to be a very small part of the whole. I can't be too much. So, I need to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
+ 314⏎Are the denominators equivalent? Yes, they are.⏎Now we can add the cardinal numbers and the numerators. First the cardinal
-
parent graph of cotangent, cot x in the figure.<|endoftext|>The steps are shown in IIS 10 on Windows 10 Enterprise, but they are also shown in IIS
-
which seems to be unique in the Indo<|endoftext|>A POTENTIALLY deadly and virulent bug that
-
Mats⏎These free blank alphabet letter mats make look simple, but they are filled with possibility for your preschool classroom! Blank ABC letters can be filled with a
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
- L0:2560 weakly excites
- L0:1518 weakly excites
- L0:2869 weakly excites
- L0:2893 weakly suppresses
- L0:2803 weakly excites
- L0:2322 weakly excites
- L0:537 weakly suppresses the 'or' in 'and/or'
- L0:1224 weakly suppresses
- L0:2360 weakly excites
- L0:827 weakly excites
- L0:1183 weakly suppresses
- L0:1190 weakly excites
- L0:437 weakly excites
- L0:1615 weakly excites
- L0:912 weakly suppresses
- L0:2808 weakly suppresses
- L0:1587 weakly suppresses
- L0:2054 weakly excites
- L0:678 weakly excites
- L0:273 weakly suppresses
- L0:2463 weakly suppresses
- L0:2454 weakly excites
- L0:178 weakly suppresses
- L0:122 weakly excites
Feeds later layers
- L9:1248 excites
- L20:386 excites
- L5:17 excites
- L15:2783 suppresses
- L21:1122 weakly suppresses
- L3:361 weakly excites
- L23:2537 weakly suppresses
- L15:576 weakly excites
- L21:1259 weakly suppresses
- L25:478 weakly suppresses
- L21:618 weakly suppresses
- L12:2824 weakly suppresses
- L13:971 weakly excites
- L3:185 weakly suppresses
- L12:2608 weakly excites
- L2:940 weakly excites
- L9:3002 weakly suppresses
- L4:918 weakly suppresses
- L12:912 weakly excites
- L8:2549 weakly excites
- L17:1783 weakly suppresses
- L7:1735 weakly suppresses
- L8:2674 weakly excites
- L3:728 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
- istics
- _CASE
- _ROUT
- 一群人
- _MISSING
- 我的心
- arer
- 口中
- 的需求
- 说我
Suppresses
- cka
- achable
- acre
- shortened
- emo
- аниз
- นำไป
- icks
- segment
- uda