'the' after 'indicate' or 'is'
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. 10 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 'the' when it comes right after verbs like 'indicate' or 'is', as in 'indicating the most job losses'. Its negative side is weaker and points at words in lists of punctuation or markup. Turning it up did not give a clear, consistent theme.
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.
-
011 to 2020. Each district within a state is given one of four distinct shades of reddish brown, with the darkest color indicating the most job losses and the lightest color showing the least. The range of job
-
called "Session Laws" or "Statutes at Large."⏎Bill Number: Bills are referred to by number. The number really has two parts: the abbreviation for the specific wing of the legislature in which the bill is introduced, as in HB (
-
escape sequences to print special characters, tabs, or line breaks Escape sequences start with a backslash ( `` ). Some common escape sequences are: - `n` Newline - Prints a newline character - `t ` Tab - Prints a tab character
-
biking to school and slow down, especially in residential areas and school zones. Motorists should know what the yellow and red bus signals mean. Yellow flashing lights indicate the bus is getting ready to stop and motorists should slow down and be prepared to stop
-
noise levels, acoustical ceiling panels should have a noise reduction coefficient (NRC) of 0.70 or greater. (The NRC indicates the average percentage of sound a material absorbs; an NRC of 0.7
-
coiled, linear and open form of DNA?⏎The equation for calculating extinction coefficients is stated by Beer's Law:⏎A = epsilon.c.l⏎Where A is the absorbance measurement at a given wavelength⏎epsilon is the extinction coefficient in M^-
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 know that, as a human, I have always been in the position of a student who has a deep appreciation for the arts, and I have always been in a position of access, and I have always been part of a community of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
in HTML. </BODY> </HTML>When this is displayed by Mosaic, it will look like:⏎As you can probably guess, commands are enclosed in angle-brackets <>, so that the HTML command <TITLE> means that the following
-
time: Servitude today?⏎(To receive these entries when they are posted, go to the upper right hand corner of the top banner and click on the three dots or parallel lines. Scroll down to the subscription form and enter your information.)<|endoftext|>
-
0 words not including the questions. Questions in bold then answers.APA 7th addition format⏎Post an explanation of the effects
-
when you have single quotation marks inside a string to avoid quote exceptions. Similarly, wrap the string in single
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
- L9:11 excites
- L7:949 weakly suppresses
- L9:900 weakly excites
- L9:3044 weakly suppresses
- L8:1007 weakly suppresses
- L8:288 weakly excites
- L2:2124 weakly excites
- L7:211 weakly excites
- L6:82 weakly suppresses
- L0:2081 weakly suppresses
- L4:2463 weakly suppresses
- L3:2179 weakly excites
- L0:2520 weakly suppresses
- L5:569 weakly excites
- L1:1827 weakly excites
- L5:2208 weakly suppresses
- L7:271 weakly excites
- L4:2116 weakly excites
- L2:867 weakly suppresses
- L8:1512 weakly excites
- L2:2232 weakly excites
- L4:735 weakly excites
- L2:759 weakly excites
- L7:1082 weakly suppresses
Feeds later layers
- L12:363 excites
- L13:2686 excites
- L13:1363 excites
- L13:2607 excites
- L11:575 excites
- L13:1420 excites
- L13:116 suppresses
- L13:2370 suppresses
- L14:1439 excites
- L12:3064 excites
- L14:1720 excites
- L15:1397 suppresses
- L13:1570 weakly suppresses
- L11:572 weakly suppresses
- L13:1191 weakly excites
- L15:2053 weakly excites
- L11:948 weakly suppresses
- L14:770 weakly excites
- L14:236 weakly suppresses 'and' joining two parallel subjects
- L14:503 weakly suppresses
- L15:1576 weakly excites
- L12:1223 weakly excites
- L11:255 weakly suppresses
- L12:460 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
- ственный
- /orders
- annon
- aste
- ;%
- unte
- instant
- instant
- ASTE
- quests
Suppresses
- ModelError
- ***!⏎
- ombine
- PropertyDescriptor
- ệt
- AndView
- 的人来说
- なりません
- tdown
- asticsearch