'after' starting a time or cause clause
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. 14 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 reacts most to the word 'after' when it introduces a clause about something that happened following another event, like 'diagnosed after 12 months'. Turning it up gave rambling text that did not clearly show this pattern.
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.
-
Bill Gates’ Breakthrough Energy Ventures. Then, in April — after recognition as one of the year’s top energy pioneers by Bloomberg New Energy Finance —
-
for peacekeeping or riot control.⏎After nearly 24 hours of flight, a Global Hawk unmanned aerial vehicle (UAV) landed successfully in South Australia after having taken off from its base in California, over 13,000
-
prompts in your own homes, kitchen tables, or where ever you create!⏎When you are finished, take a photo and tag @PlainsArtMuseum on Instagram or Facebook
-
it’s at rest.⏎It’s best to check it in the morning after you’ve had a good night’s sleep and before you get out of bed. The average
-
, and everything from flowers to fresh fruit in short supply across northern Europe, most people just wanted to get things moving again. Late on April 20, after six full days—the longest, most expensive flight delay in history (the closure of
-
uly burdensome or discriminatory land use regulations. The law was passed unanimously by Congress in 2000, after hearings in which Congress found that houses of worship, particularly those of minority religions and
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 think of, now again, and then in the present day, as well as now and then, and then again, as well as in the past, and then... etc. and then... as well, etc. and then,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Struggle-For-Power-In-The-994323.html⏎"The Struggle for Power in the Ussr" StudyMode.com. 05 2012. 05 2012
-
⏎The main crack<|endoftext|>analysis of algoritms⏎robert.kern at gmail.com⏎Fri Sep 10 00:10:22 CEST 2010⏎On 9/9/10
-
of certain drugs, or as in inherited disorder.⏎Use the citation below to add this definition to your bibliography:⏎"acanthosis nigricans." Definitions.net. STANDS4 LLC, 2013. Web. 2
-
2 km. The infrasound data is collected at 20 hz, while the meteorologic data is recorded at 1 hz.⏎Figure 4. Array layout. Blue dots indicate the 7 elements, while the purple
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
- L13:908 suppresses
- L9:1727 excites
- L10:775 suppresses
- L12:2720 suppresses
- L5:1831 suppresses
- L13:2809 weakly excites
- L11:2945 weakly suppresses
- L13:2922 weakly excites
- L8:1551 weakly excites
- L8:1946 weakly excites comma after 'who' or 'which'
- L12:109 weakly excites
- L1:1517 weakly suppresses
- L7:1239 weakly suppresses
- L13:1874 weakly excites
- L11:1768 weakly suppresses
- L12:128 weakly excites
- L13:91 weakly excites
- L1:2936 weakly suppresses
- L4:2485 weakly excites
- L6:2035 weakly excites
- L4:1925 weakly excites
- L11:698 weakly excites
- L7:737 weakly suppresses comma after a place name
- L4:309 weakly excites
Feeds later layers
- L16:1153 excites
- L17:1512 weakly excites
- L18:2358 weakly suppresses
- L19:1970 weakly suppresses
- L17:2646 weakly excites
- L18:1378 weakly excites
- L19:2705 weakly excites
- L22:1841 weakly suppresses
- L27:2909 weakly suppresses
- L16:3022 weakly excites
- L18:1710 weakly suppresses
- L17:2292 weakly excites
- L15:622 weakly excites
- L20:2367 weakly excites
- L24:712 weakly excites 'long' in 'long-term' and 'long-lasting'
- L23:947 weakly excites
- L22:1465 weakly excites
- L16:2796 weakly suppresses
- L19:1712 weakly excites
- L24:1839 weakly excites
- L16:1890 weakly suppresses
- L24:2923 weakly excites
- L19:2603 weakly excites
- L24:1394 weakly 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
- carefully
- 漫长的
- doInBackground
- bitter
- 一行
- 阉
- successfully
- AIN
- mess
- ään
Suppresses
- ounded
- /goto
- aujourd
- aise
- merican
- egin
- cci
- alar
- fähig
- aaS