question mark ending a web question heading
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 question mark at the end of a question-style heading, the kind often seen in FAQ pages and article titles, right before the answer begins. Turning it up gave off-topic or number-filled text, so the steering does not show this pattern.
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.
-
You might also be interested in:⏎What do smog, acid rain, carbon monoxide, fossil fuel exhausts, and tropospheric ozone have in common? They are all examples of air pollution.<|endoftext|>Revealing the art and architecture
-
culture.⏎This resource is licensed by LoubeeSav⏎under the iSLCollective Copyright License<|endoftext|>|Does a 'proper' tragic hero need a crown?|⏎Christmas Eve brought an email inviting me to write an essay about tragedy and social
-
selection of participants⏎- Type of participants⏎- democratic innovation yields a non-binding decision<|endoftext|>What does ‘diff’ mean in Git? A diff takes two data sets and shows you what has changed between them. Data
-
<|endoftext|>Which Joe gave his name to ‘sloppy joes’? We look at five interesting sandwiches and their lexical origins.⏎A small North American diving
-
now have customized climate predictions, thanks to a<|endoftext|>What is redistricting and how does it work in Iowa? Your questions answered.⏎The U.S. Census Bureau delivered its long-delayed population
-
or public awareness can be developed;⏎- Economic and Cultural Issues – Most of the efforts<|endoftext|>Doesn’t it seem like food allergies are way more common these days? When I was a kid, nobody worried about serving peanut butter at school, and
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:::::::::::::::::::::::::::::::::::::::::::::::::
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
berg says.⏎Think fat just hangs around and does nothing? It doesn’t⏎By Nanci Hellmich, USA TODAY⏎PHOENIX — Most people think of fat as an inert blob, but fat cells release powerful
-
, more than 3,800 among adults 55 to 64 and around 800 among adults older than 65 — they accounted for a smaller proportion of deaths in these populations.⏎Among the eldest
-
, 2012⏎Share Email Print⏎Kate Wong is an editor and writer at Scientific American covering paleontology, archaeology and life sciences. Follow on Twitter⏎PALEO DIET: Analyses of tartar on the teeth
-
All intellectual property rights to the game are owned by respective owners in the U.S.A and Canada and the rest of the world. Anagrammer.com is not affiliated with Scrabble. This site is an educational tool
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
- L18:2582 excites
- L18:5 excites
- L17:829 weakly excites
- L4:221 weakly suppresses
- L18:2043 weakly excites
- L14:392 weakly suppresses
- L2:204 weakly suppresses
- L6:718 weakly excites
- L16:32 weakly excites
- L2:7 weakly excites
- L18:887 weakly excites
- L16:118 weakly suppresses
- L9:7 weakly excites
- L17:38 weakly suppresses
- L14:51 weakly excites colon that introduces a statement
- L9:1109 weakly suppresses
- L18:2455 weakly excites
- L18:106 weakly excites
- L18:429 weakly excites
- L18:156 weakly excites
- L6:2486 weakly excites
- L14:1067 weakly excites
- L6:2678 weakly suppresses
- L9:366 weakly excites
Feeds later layers
- L21:189 suppresses document break token mid-sentence
- L20:1313 suppresses
- L22:541 excites
- L26:130 weakly excites ellipsis cut-off in quotes
- L25:273 weakly excites dollar sign before an amount
- L27:1053 weakly suppresses
- L22:416 weakly excites commas in dates and number lists
- L20:353 weakly excites
- L27:111 weakly suppresses
- L20:79 weakly suppresses
- L24:94 weakly suppresses
- L23:35 weakly excites
- L20:777 weakly suppresses
- L27:979 weakly suppresses
- L23:1878 weakly suppresses
- L20:398 weakly excites
- L25:360 weakly excites
- L26:56 weakly excites
- L24:2482 weakly suppresses
- L20:1710 weakly suppresses
- L21:1086 weakly excites end of sentence before a claim or statistic
- L21:50 weakly excites comma after 'So' or intro phrase
- L20:59 weakly suppresses
- L20:346 weakly suppresses
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
- cup
- lse
- ��
- ///<
- .toHexString
- כן
- 实
- `,
- 轿
- 庾
Suppresses
- OMATIC
- unsch
- ""),⏎
- lessly
- -hash
- arine
- eds
- esh
- LOT
- -rich