blank slot before a year or number
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. 26 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 an empty spot where a number, usually a year, is missing, such as 'since the [blank]1950s'. Turning it up made the model produce strange filler text and repeated words.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
belly. This procedure is generally performed under general anesthesia in a hospital.⏎- MRI. This test uses a large magnet, radio waves, and a computer to make 2-D images of an internal
-
investors. This warning contradicts the widely held belief that global crude reserves will remain plentiful for decades.⏎Critics say similar predictions of scarcity at the time of the 1973-74 Arab oil embargo didn’t come true. “
-
pain or discomfort, or a severe cough, contact your health care provider, walk-in clinic, or emergency department. If you are having a medical emergency, call 9-1-1.⏎Tips to reduce your smoke exposure⏎Smoke levels may
-
. The Washington Post’s Robert Samuelson has done just this in a recent column. As pointed out by Mr. Samuelson blacks have made much progress since the 1960
-
/an untrained eye<|endoftext|>From SABR member John Thorn at Our Game on August 18, 2014:⏎When Jackie Robinson opened the 1947
-
began a collaboration with DeepMind and the team behind AlphaFold - a machine learning program that attracted significant attention from the structural biology community based on its performance in the 14thCritical Assessment of<|endoftext|>Researchers Show Impaired Muscle Function from Antibacterial
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: (心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先心中先
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
now accepted defeat.⏎The brief, fragmentary Book <|endoftext|>Welfare state in capitalism⏎Since the eruption of the sub-prime mortgage crises of 2007 to 2009, there have been hot debates on how the
-
, fresh produce may be secured. The home vegetable garden is worthy of<|endoftext|>MicroRNAs (miRNAs) are single-stranded, 18–24 nucleotide long, non-coding RNA molecules. They are involved in virtually
-
18mm lens (because I wanted to have as much of the sky as possible in my image) my equation was 600/18=33 seconds.⏎With an exposure of 33 seconds I had to have a
-
that ignites spontaneously in air when finely divided. Strontium is used in pyrotechnic compounds and various alloys. Atomic number 38; atomic weight 87.62; melting point 76
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
- L25:146 strongly excites
- L25:854 strongly suppresses
- L22:743 strongly suppresses
- L21:395 strongly excites
- L24:197 strongly excites
- L25:232 strongly suppresses
- L22:2275 strongly excites
- L25:182 strongly excites
- L23:199 suppresses
- L21:580 suppresses
- L25:3007 suppresses the zero digit in decimals
- L19:298 excites
- L23:416 suppresses
- L20:787 excites
- L22:193 suppresses
- L19:1985 excites
- L24:13 excites
- L21:110 excites
- L22:332 suppresses
- L25:95 suppresses
- L20:1337 suppresses
- L21:339 excites spot just before a year
- L24:162 suppresses
- L16:372 excites
Feeds later layers
- L27:238 strongly suppresses
- L27:48 strongly excites
- L27:241 strongly excites
- L27:283 strongly suppresses the spot right before a number
- L27:428 suppresses
- L27:530 suppresses
- L27:628 suppresses
- L27:396 weakly suppresses
- L27:328 weakly suppresses
- L27:287 weakly suppresses
- L27:448 weakly excites
- L27:1456 weakly suppresses
- L27:132 weakly excites
- L27:22 weakly suppresses
- L27:1019 weakly suppresses
- L27:494 weakly suppresses
- L27:1332 weakly suppresses
- L27:1132 weakly excites
- L27:2249 weakly suppresses
- L27:154 weakly excites
- L27:913 weakly suppresses
- L27:1196 weakly suppresses
- L27:323 weakly suppresses
- L27:1113 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
- ㄒ
- Authorities
- URLException
- "</
- URLOPT
- StateException
- Ḋ
- /{{$
- HTTPHeader
- (${
Suppresses
- 1
- 2
- 3
- 5
- 8
- 4
- 9
- 6
- 0
- 7