'of' in 'As of' dates
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. 1 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 to the word 'of' in the phrase 'As of', which is usually followed by a year or a census figure. Turning it up did not produce any clear change in the model's writing that relates to this.
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.
-
state of Utah. As of the 2010 census, the population was 231,
-
m²)⏎|Time zone||Central: UTC-6/-5|⏎Cherokee County is a county located in the U.S. state of Texas. As of the 2010 census, its population was 50,8
-
50 percent of electricians made between $37,570 and $65,260 a year.⏎Salary by Employment Sector⏎As of 2011, the highest-paid electricians were those who taught at
-
a high school diploma, but all electricians must undergo an apprenticeship of up to four years before they can become licensed.⏎Average and Median Earnings⏎As of 2011, electricians earned an average of $52,
-
Alberta generated $21.3 billion in revenue.⏎- Alberta has the third largest food and beverage manufacturing sector in the country after Ontario and Quebec.⏎- As of 2020, there were over 82,000
-
following a public health order, enforcement will rely on education and individual responsibility for the good of the community as a whole.⏎“There have been reports<|endoftext|>Note: As of the 2022-2023 program year, the Cyber
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 way I was supposed to be. I am a bit of a student. I need to know this. I need to learn. I am a bit of a student. I need to know this. I am a bit of a student
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
as possible. For the 2012 hurricane season, the industry continues to build
-
liberated city, should we turn the page as soon as possible to erase the devastation wrought by ISIS? Rebuild no matter the cost, with
-
capacity to move water from affected areas as quickly as possible out to tide, to the Water Conservation Areas or into Lake Okeechob
-
a result of government policy.⏎The goal is "to narrow as soon as possible the gap in economic development between Tibet and other areas of the nation" (White
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
- L0:1301 suppresses
- L0:1224 suppresses
- L0:178 suppresses
- L0:1075 excites
- L0:1355 weakly excites
- L0:1 weakly suppresses
- L0:2420 weakly suppresses
- L0:793 weakly excites
- L0:2042 weakly suppresses
- L0:2726 weakly excites
- L0:852 weakly excites
- L0:1579 weakly suppresses
- L0:761 weakly suppresses
- L0:633 weakly excites
- L0:93 weakly suppresses
- L0:563 weakly suppresses 'www' in web addresses
- L0:1232 weakly excites
- L0:2819 weakly suppresses
- L0:2684 weakly suppresses
- L0:2389 weakly suppresses
- L0:191 weakly suppresses
- L0:2201 weakly suppresses
- L0:2068 weakly excites
- L0:2320 weakly suppresses
Feeds later layers
- L8:495 weakly excites
- L13:2158 weakly excites
- L9:440 weakly excites
- L4:522 weakly excites
- L14:2132 weakly suppresses
- L4:724 weakly excites
- L23:1378 weakly suppresses
- L6:182 weakly suppresses
- L4:1060 weakly excites
- L19:1399 weakly suppresses
- L14:633 weakly excites
- L12:295 weakly suppresses
- L5:20 weakly excites
- L6:1455 weakly excites
- L11:1189 weakly excites
- L24:851 weakly excites
- L3:343 weakly excites
- L4:117 weakly suppresses
- L21:1049 weakly excites
- L2:215 weakly suppresses
- L17:451 weakly suppresses
- L24:1085 weakly suppresses
- L17:2537 weakly suppresses
- L22:2136 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
- 他们是
- rops
- heck
- hk
- heimer
- 他们在
- ental
- ichten
- 他们的
- sess
Suppresses
- рид
- elman
- olta
- 酚
- Ref
- Kind
- lef
- Geoff
- insulated
- 主动