the word 'this' after a preposition
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. 15 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 responds to the word 'this' when it follows a preposition, as in 'using this form', 'see this article' or 'buy this book'. It is pushed the other way by words like 'That', 'This' or 'These' that start a new sentence. Turning it up or down did not give steered text that clearly relates to this word.
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.
-
27.5 West). 38 meter/pixel resolution.⏎Note: this THEMIS visual image has not been radiometrically nor geometrically calibrated for this preliminary release. An empirical correction has been performed to remove instrumental effects. A linear
-
Android mobile device you will need the Adobe Digital Editions app, or BlueFire Reader or Txtr app. These are free, too. For more details see this article.⏎|Size: ||359 KB|⏎|Date published: ||
-
What professor is this problem relevant for?⏎Based on our data, we think this problem is relevant for Professor Finnegan's class at WSU.⏎What textbook is this problem found in?⏎Our data indicates that this problem or a close variation was asked
-
10th - 12th English Language Arts⏎Lesson Plan: The State of the Union⏎Learners explore the state of the union. In this U.S. government lesson, students analyze the 1998 State of
-
Android mobile device you will need the Adobe Digital Editions app, or BlueFire Reader or Txtr app. These are free, too. For more details see this article.⏎|Size: ||4.4 MB|⏎|Publisher: ||CRC
-
, Butler, Clark, Clermont, Delaware, Franklin, Greene, Hamilton, Montgomery and Warren.⏎Dead birds should be reported to the Ohio Division of Wildlife using this form.⏎"Kentucky Fish and Wildlife has received 2,300
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: this is that "something that which that that that that" that that that that that that that that that that that that that that that that that that that that that that that that that that that that that that that that that that that that
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
in each of our cells that holds the architectural plan of our body and physical functions. Over the generations, DNA changes at a presumably steady rate. For birds, that's about 2% change (for mitochondrial DNA) in a million years --
-
are many reasons to switch to solar mobile power sources. The first reason is that solar energy is a renewable resource. This means that it will never run out and we can rely on it for our energy
-
blood vessels become narrow, blood flow tends to slow down and form clots. These clots may get dislodged and block other blood vessels. Depending on the
-
success for domestic rubbish. This is because the biomass in domestic rubbish can only be gasified at low temperatures, but this in turn creates tars and waxes that stifle the process.⏎However,
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
- L4:968 excites
- L10:965 suppresses the word 'that' introducing a clause
- L4:468 excites "that" after "than" or "like"
- L12:521 excites
- L9:82 suppresses
- L0:1035 excites
- L10:167 suppresses
- L11:199 excites
- L6:1080 excites "that" as a pointing word
- L6:1218 suppresses
- L13:112 suppresses
- L1:1318 excites
- L4:270 suppresses
- L14:2364 weakly suppresses
- L8:10 weakly excites
- L14:1807 weakly suppresses
- L14:332 weakly excites
- L3:1004 weakly excites
- L5:125 weakly excites "of" after all three, both, ten
- L12:51 weakly suppresses
- L12:336 weakly excites
- L12:431 weakly excites
- L1:2290 weakly suppresses
- L3:1068 weakly excites "that" pointing back to a noun
Feeds later layers
- L17:2011 strongly suppresses
- L16:2466 strongly suppresses
- L17:221 strongly suppresses
- L16:2861 suppresses
- L19:1737 suppresses
- L17:1816 suppresses
- L19:2547 suppresses
- L20:1175 suppresses
- L17:390 suppresses
- L18:1422 suppresses
- L19:1341 suppresses "this" after a time phrase
- L18:305 suppresses
- L16:815 suppresses 'the' before a noun mid-sentence
- L18:1944 suppresses
- L16:266 excites
- L19:1543 suppresses
- L25:926 suppresses
- L17:1022 suppresses
- L22:2625 suppresses
- L20:1601 suppresses
- L16:2660 suppresses
- L19:2958 suppresses
- L17:510 suppresses
- L20:866 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
- ivals
- ourage
- nees
- 역
- -thumb
- >?
- 椰
- arger
- Reflex
- coc
Suppresses
- section
- 过程
- kind
- process
- 方面的
- enta
- happening
- 家伙
- tering
- ഹ