the digit 5 inside numbers
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. 19 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 the digit 5 when it sits in the middle of a number, such as in a decimal, a DOI or a range like 33.0 to 35.6. It is not a clear picture of a broader idea, and the steering results do not show a consistent theme.
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.
-
AAAJMR-22-82571(R); Published: 29-Nov-2022, DOI:10.35841/ aaajmr-6.11.155
-
company insiders.⏎IBM, who produced the figures based on information from over 8,000 of their clients devices, revealed that although 15.5 percent of such ‘attacks’ were caused inadvertently, 44.5 percent
-
greater proficiency, they may invest in a drum. Generally diameter is between 13 and 14 inches (33.02-35.56cm). Depth of the drum, as measured
-
553 or BIO558. This course is an accepted elective in our MAT program, our Masters in
-
analyzer is a display unit with a battery of filters centered (if it is "third of octave" model) on 25, 31.5, 40, 50, 63, 80
-
,” Christina Gibson-Davis. Demography, July 2, 2014. DOI: 10.1007/s13524-014-0308-7<|endoftext|>Designs
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 to be more than just the most beautiful as a person, but also have to be more than just the most beautiful as a person's ideal. I have to have the most beautiful as a person, but also have to be
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the electron charge, and they apparently couldn’t be observed anyway), why would you?⏎Of course, we now<|endoftext|>Logical flaws responsible for complex evolutionary theory⏎|May 21, 2011||Posted by idnet.com.au
-
00-lb load would use the following calculation:⏎[(4,000 lbs + 6,000 lbs) × 1.25] ÷ 4 = 3,125 lbs⏎According to
-
this post answers some of those questions!⏎IDEA Public Schools, Coming Soon to Southern Louisiana, is a Public Charter School Network—What Does That Mean?⏎You may have heard, that in 2018, IDEA Public Schools
-
sweet gum tree and more.⏎Now, I invite you to make this journey.⏎Narrow body with pointy edges⏎Narrow body with pointy edges⏎May grow
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
- L18:778 strongly suppresses
- L4:646 excites
- L0:1811 excites
- L2:1925 excites
- L10:1501 excites
- L16:249 excites
- L16:37 excites
- L4:969 suppresses the letter C in vitamin C
- L17:241 excites
- L15:89 excites
- L15:78 excites
- L15:80 suppresses
- L15:129 suppresses
- L15:115 excites
- L14:643 excites
- L18:440 suppresses
- L16:113 suppresses
- L4:2938 weakly suppresses
- L18:402 weakly excites
- L16:184 weakly suppresses
- L15:96 weakly suppresses
- L18:88 weakly suppresses
- L17:337 weakly suppresses
- L16:99 weakly suppresses
Feeds later layers
- L20:2681 suppresses
- L21:1979 suppresses the number 5 in a numbered list
- L24:1676 suppresses zeros in large round numbers
- L22:3071 excites
- L20:2255 excites
- L20:1085 excites
- L20:931 weakly excites the digit 6 inside years
- L21:2676 weakly excites
- L20:558 weakly excites
- L25:2807 weakly suppresses
- L21:308 weakly excites
- L24:124 weakly excites dollar sign before a money amount
- L21:723 weakly excites
- L26:1355 weakly excites
- L22:2379 weakly excites
- L20:2217 weakly excites 'trying' after 'were' or 'are'
- L23:900 weakly suppresses
- L20:1374 weakly excites
- L20:416 weakly excites the "1" in years 191x
- L24:2840 weakly excites
- L20:2045 weakly excites
- L20:1308 weakly excites the digit 3 inside years
- L24:2097 weakly suppresses
- L23:2739 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
- Representation
- spring
- 露
- 不停
- 'Neill
- ortion
- ula
- 室
- emple
- 存在
Suppresses
- -HT
- divider
- +-+-
- igram
- agnar
- nam
- uits
- haps
- cruiser
- /archive