periods ending informational sentences
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. 18 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 lights up on the full stop at the end of a plain informational sentence, like the end of a sentence explaining a topic in an article. When it was turned up, the model's text drifted into rambling repetition, and when turned down it tended toward quiz-style answer choices.
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.
-
growth⏎Coffee is exclusively beneficial for hair health<|endoftext|>Ninjas vs Superbugs⏎Antibiotic-resistant bacteria - or, superbugs - are a growing threat. And MRSA is one of the worst. Learn more about Ninja Polymers
-
tgz File Using the Terminal?⏎If you are using open-source packages, then sooner or later, you will encounter the .tgz or .tar.gz files. This is a variation of zipped files; most open-source packages use this format
-
on the cross tabulation feature in Survey Analytics. Cross tabulations are a mainstay in the market research industry and is one of the most widely-used statistical tools. Cross
-
sports, in the NERF Camp⏎Probably one of the most fun activities where little ones can do a lot of movement playing outdoors is NERF camp. Here, children have the opportunity to play and bond with friends with other 3
-
most popular cryptocurrencies which can be used in 2020.⏎Maybe you’ve heard the terms digital gold or halving if you’ve been listening close enough. More importantly, you may have even heard about how Bitcoin or BTC prices have been
-
and stroke, among others using the same procedure.⏎MRI scans are painless and, unlike X-ray scans, they avoid X-ray radiation exposure, notes MedicineNet. There are
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 key point, as for the number, is to make the correct number. So, when I have been trying to think about the correct number, it is clear that the statement, "x" for the number, is the correct one
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
human activities can theoretically exert some influence.(7) As the 21st century began, some experts thought it plausible
-
collection of wrought iron rivets, saddles and armature bars that make the statue structurally sound, and a double helix metal staircase is centered among them./Jared Owen<|endoftext|>New Study Issues Big Warning On Global Obesity Rates…⏎Published in
-
, freedom for the oppressed, and abundance for the poor. Therefore, the people waited for the Messiah or Christ.⏎The word Christmas, or "Christes Maesse," was first used in
-
labor. This absence of any higher comprehensive social philosophy led to efforts to supplement it by providing an overall philosophy of the public good. This absence led to Progressivism.At first it was a diverse movement of many who thought of themselves as “progressives
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
- L17:944 suppresses
- L16:489 suppresses
- L15:2341 suppresses
- L17:2017 excites
- L17:1426 suppresses
- L2:156 excites
- L5:3 suppresses
- L9:12 suppresses
- L4:7 excites
- L14:1187 suppresses
- L9:2447 excites
- L16:2857 excites
- L17:990 excites
- L9:7 excites
- L15:1260 excites
- L16:1112 weakly suppresses
- L1:130 weakly suppresses
- L8:3048 weakly excites
- L10:33 weakly suppresses
- L12:731 weakly excites
- L15:2811 weakly excites
- L16:432 weakly suppresses period after month or initial
- L1:1 weakly excites
- L9:103 weakly excites "as" or "since" giving a reason
Feeds later layers
- L20:1147 excites comma after a sentence subject
- L20:1799 weakly suppresses
- L24:1720 weakly excites
- L20:2998 weakly suppresses
- L27:2653 weakly suppresses
- L19:1105 weakly suppresses
- L19:1485 weakly suppresses
- L19:2548 weakly suppresses
- L23:1875 weakly suppresses
- L26:1141 weakly excites
- L26:2170 weakly excites
- L19:297 weakly suppresses
- L21:555 weakly excites
- L21:3059 weakly suppresses 'the' after words like 'underscores'
- L24:3070 weakly suppresses
- L19:2370 weakly excites
- L27:1182 weakly excites
- L20:448 weakly excites
- L19:1656 weakly excites
- L25:2942 weakly excites
- L22:688 weakly suppresses
- L24:2307 weakly excites
- L20:1806 weakly excites slash inside web addresses
- L26:189 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
- ˗
- помощи
- |m
- acking
- }}/
- <Action
- 付费
- #undef
- ặng
- 掩
Suppresses
- ⏎⏎⏎
- ⏎ ⏎
- aroo
- .
- entre
- Grad
- ta
- Liqu
- scious
- sc