'for' introducing a purpose or target
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. 16 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 'for' when it points to a purpose or recipient, like 'eBook for Free', 'activities for the book', or 'for seven days'. Its negative side fires on 'there' in 'as many ... as there are'. Steering output was unrelated text, so the pattern rests on the snippets alone.
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.
-
In this report, the authors conclude with implications for policymakers and LEAs.⏎Challenges in Implementing Physical Security Measures in K–12 Schools⏎Download eBook for Free⏎|PDF file||0.6 MB||⏎Use Adobe Acrobat Reader version
-
a perfect enhancement to E.L. Konigsburg's Newberry Medal Winner.⏎This unit contains vocabulary, comprehension and writing activities for the book, The Magic School Bus: a Hurricane by Joanna Cole. Common Core
-
, if errno is non-zero, an error has occurred. The ldexpf() and ldexpl() functions do not set errno.⏎See attributes(5) for descriptions of the following attributes:<|endoftext|>The
-
you will be asked to critically think and debate varying interpretations of the empirical evidence in relation to the science being brought to your attention<|endoftext|>Economics– Important Questions Bank for Rajasthan Board HSC 201
-
recommend or endorse any third party, product, service or information provided on this site.⏎If you have dietary restrictions and/or allergies, always read the ingredient list carefully for all products prior to consumption. If the ingredient list is not available on the product
-
. Nicklin.”⏎Israel Thomas Nicklin was born in Loudon County, Virginia, August 12, 1797. After attending Fairfax School for several years, he moved with his parents to Bowles Mills, Tyler County,
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 first's, first's, and first's. I think it's going to be a first's. Maybe's. Maybe's. Maybe's. So, maybe's. But, maybe's. Maybe's. Wait, maybe's
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the pinnacle and its prize, only to lose their footing and slide back down to their harm or death.⏎How many glass mountains are there? Maybe as many as there are people. For me, the most devious of glass mountains appears in the
-
are 221293000 people.⏎Pakistan telephone code is 92.<|endoftext|>T HERE are as many definitions of public opinion as there aredisciplinary points of view. Political scientists see public opinion interms of the distribution
-
⏎Review: Brave. A Book About Courage⏎There are as many definitions of poetry as there are poets. Poetry is an artistic expression unlike any other. Some research suggests that
-
residence itself, or a brook falling down the sward between trees and bushes and clumpy growths. There are as many forms and kinds of gardens as there are persons who have gardens […].” –LHBGC, p.
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
- L5:858 strongly excites
- L1:47 excites
- L2:1866 suppresses
- L0:41 suppresses
- L1:32 excites
- L0:2724 suppresses
- L3:381 excites 'for' in research institute names
- L10:1473 weakly suppresses
- L15:1587 weakly excites
- L2:1925 weakly suppresses
- L15:792 weakly suppresses
- L0:1543 weakly excites
- L0:564 weakly excites
- L0:2765 weakly excites
- L3:1055 weakly suppresses
- L4:189 weakly excites
- L15:1135 weakly suppresses
- L11:2595 weakly suppresses
- L15:1616 weakly excites word fragments mid-word
- L1:2601 weakly suppresses
- L15:201 weakly excites
- L12:2810 weakly suppresses
- L15:1900 weakly suppresses
- L0:2211 weakly suppresses
Feeds later layers
- L18:650 suppresses the word 'more' before a comparison
- L27:2224 suppresses
- L17:253 suppresses
- L19:572 suppresses
- L20:601 suppresses
- L21:812 suppresses
- L19:1070 suppresses
- L19:1237 suppresses
- L19:1382 excites 'addition' in 'In addition to'
- L23:1368 suppresses
- L20:934 excites
- L26:1286 suppresses
- L21:2717 excites
- L22:2883 suppresses
- L18:778 excites
- L17:1928 excites
- L19:1728 weakly suppresses
- L19:2221 weakly excites
- L26:2860 weakly suppresses
- L22:1141 weakly excites
- L20:2375 weakly excites
- L26:2583 weakly suppresses
- L18:687 weakly excites
- L25:2511 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
- sg
- e
- 冲动
- \xd
- \xb
- olatile
- ingular
- escription
- arrow
- ⎝
Suppresses
- iously
- ium
- bij
- ialis
- liable
- 镡
- anium
- thur
- rett
- izont