last word before end of document
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. 19 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 final word or punctuation mark just before a text ends, where the next token is the end-of-document marker. The steering results are mostly unrelated, with repeated dots, numbers or stray words, so they do not add much.
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.
-
takes longer than mechanical disruption.⏎Can't find the product you've used to perform this experiment? It would be great if you can help us by Adding a product<|endoftext|>In 1800s, structural steels have started to use in
-
460 degrees Celsius<|endoftext|>For the Union Dead⏎by Robert Lowell⏎Stanza 10 Summary⏎
-
of balloons.⏎Submit a review<|endoftext|>1 year ago⏎Verified Sherpa Tutor⏎If a liquid has a definite volume
-
officer dies in the line of duty.⏎- The peace officer mourning ribbon does not take the place of the poppy on Remembrance Day.⏎- Date modified:<|endoftext|>Research mines to discover minerals and the best ways to get them out.⏎Engineers
-
limestone and cullet⏎Information provided by the Glass Packaging Institute http://www.gpi.org.⏎Answers: C, A, B, B, C<|endoftext|>This article is about the Great American Novel as a concept. For other uses,
-
how the common language runtime uses the machine configuration file for assembly binding, see How the Runtime Locates Assemblies.⏎An application configuration file contains settings that are specific<|endoftext|>Mercury is used in thermometers, fluorescent light bulbs, and some electrical switches
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 am a 30 years old man, 3 years ago I was 1
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
composition. The basic⏎source of all minerals is the hot magma in the⏎interior of the earth.⏎(more content follows the advertisement below) A D V E R T I S E M E N T⏎When magma cools,⏎
-
Extended Library Functions, Volume 2 Oracle Solaris 11.1 Information Library|⏎- load exponent of a floating point number⏎c99 [ flag... ] file... -lm [ library... ] #include <math.h> double
-
the refracted angle ( r ).⏎- The light beam can bend towards the normal or away from the normal.
-
the outer loop test in step 4 will execute ( n + 1 )⏎> times (note that an extra step is required to terminate the for loop
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
- L18:795 suppresses
- L18:246 suppresses
- L16:795 suppresses
- L16:317 excites
- L18:955 excites
- L17:15 suppresses
- L6:2543 excites
- L18:48 weakly suppresses
- L9:234 weakly suppresses line break inside hard-wrapped text
- L18:2566 weakly excites
- L7:1240 weakly suppresses
- L16:455 weakly excites
- L10:527 weakly excites
- L17:2095 weakly suppresses
- L0:2059 weakly excites
- L0:22 weakly excites
- L18:76 weakly suppresses blank spot before a number
- L9:1768 weakly suppresses
- L17:1568 weakly excites
- L11:20 weakly excites
- L16:170 weakly suppresses
- L6:2157 weakly excites
- L9:2943 weakly excites
- L16:2945 weakly excites
Feeds later layers
- L22:16 strongly suppresses comma in dates and birth lines
- L26:21 strongly suppresses
- L20:59 strongly excites
- L27:11 strongly suppresses unclear: mixed fragments in lists and pronunciations
- L27:33 strongly excites
- L25:109 suppresses
- L22:14 excites
- L24:39 excites
- L25:17 excites
- L27:97 excites
- L21:418 excites
- L24:51 excites
- L20:489 excites
- L24:33 suppresses
- L25:30 excites
- L23:80 excites blank slot before a year or number
- L21:70 excites
- L27:101 excites
- L20:193 suppresses
- L23:60 excites
- L26:103 excites blank space before a number
- L23:19 suppresses
- L27:71 suppresses
- L24:94 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
- igor
- adal
- ibus
- éis
- -strokes
- atos
- they
- aday
- 循
- 性的
Suppresses
- "
- L
- '
- “
- ‘
- US
- A
- +
- UN
- American