the gap in number ranges
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 fairly consistent: several of them 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 spot between two numbers in a range, like the gap in '10-15', '50 to 100' or '12 to 18 inches'. It is most active right before the second number of the range is about to appear. When turned up the model produced repeated words or long strings of zeros, so the steering output does not show what the neuron does.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
quite variable, but a reasonable mean is about 10-15 from each protein. They all have to agree in their "chemical properties
-
this, he proposed a machine that can collect data, compiling it into a type of artificial intelligence.⏎"I think as we look toward the next 50 to 100 years, if we really can come up with these machines that can
-
't known how long they can live under professional care, she said, because zoos and aquariums have only tracked animal ages for the last 30 to 40 years.⏎Thunder, who was most recently seen in the Canyon Falls exhibit
-
risk for type 2 diabetes, heart disease, osteoarthritis, and certain forms of cancer.⏎âIf youâre overweight, losing just 5 to 10 percent of your body weight can slash the risk of both type
-
torso in horizontal line, rotating shoulders. Head and shoulders stay aligned so they move as single unit. Partners reverse sequence and return to start, performing 10–15 reps before switching sides.⏎Medicine Ball Partner Twists⏎Partners
-
full sun and well-drained soil. Prepare the garden bed by using a garden fork or tiller to loosen the soil to a depth of 12 to 15 inches, then mix in a 2- to 4-inch layer
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 have, in my life, as, and, and, and, and, and, and, and, and, as, and, and, and, and, and, and, and, and, and, and, and
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
€™s WONDER database, which tracks mortality data and causes of death in the U.S. Researchers first isolated all opioid-related deaths recorded between 2001 and 2016 (335,1
-
Large Scale Biosphere-Atmosphere|⏎|Time Period||1985-01-01 to 2001-12-
-
6 and ES-3, tbl. 7, with EPA, Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990-2019<|endoftext|>Following directions with color words is a fun practice activity
-
.⏎Clinics that gave the suspect shots<|endoftext|>Classic DACB CollectionAll articles created or submitted in the first twenty years of the project, from 1995 to 2015.⏎Jacob Osasona was born in the
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
- L13:1435 excites
- L11:2354 suppresses
- L12:947 suppresses
- L5:1401 weakly excites
- L13:2019 weakly excites
- L7:1643 weakly suppresses
- L12:1965 weakly suppresses
- L11:2811 weakly suppresses
- L1:2559 weakly excites
- L13:2379 weakly suppresses the digit zero in numbers
- L13:707 weakly suppresses
- L8:932 weakly suppresses
- L9:850 weakly suppresses
- L5:778 weakly suppresses
- L10:2889 weakly excites
- L8:2001 weakly excites
- L8:1946 weakly suppresses comma after 'who' or 'which'
- L11:1878 weakly excites
- L3:2724 weakly suppresses
- L8:564 weakly suppresses
- L8:364 weakly suppresses
- L7:1429 weakly excites
- L6:1577 weakly excites
- L14:1420 weakly suppresses
Feeds later layers
- L16:808 strongly excites
- L18:1413 excites
- L18:2454 excites
- L17:483 suppresses
- L18:2238 excites
- L18:1207 excites
- L21:1972 excites
- L17:2704 excites
- L19:1393 excites
- L16:548 excites
- L21:1857 excites
- L18:2362 excites
- L19:2835 excites
- L19:659 excites
- L23:223 excites
- L22:1248 excites
- L17:1468 excites
- L19:2735 excites
- L25:41 suppresses
- L27:376 excites
- L20:1009 excites
- L17:2116 excites
- L20:560 excites
- L20:742 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
- 或
- or
- 或者是
- または
- 或多
- 或
- 或是
- hoặc
- /she
- 或者
Suppresses
- chair
- rek
- ums
- ceu
- rix
- uner
- enade
- 法
- 법
- AMPLE