ampersand joining two items
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 lights up on the "&" symbol when it joins two things in a title or list, like "Backstage, & Offstage". It goes quiet or negative on the word "and" in phrases like "by and large". Turning it up made the output wander into unrelated text, so the steering doesn't reveal much.
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.
-
by Dunn & Price)⏎Also see posts: “Adult Learning styles“⏎Existential – these thinkers are focused on issues too deep for other kids to understand, & may be prone to depression as they ponder the meaning of life. Careers involving philosophy
-
lesson plans or texts, but in schematic form, suggesting avenues along which the conversation might develop.<|endoftext|>Introduction to Production⏎Creating Theatre Onstage, Backstage, & Offstage⏎Introduction to Production: Creating Theatre Onstage, Backstage, &
-
, courgettes⏎3. WHEN? (see crop chart)⏎Right time: temperature, how long til harvest? Seed packets will tell you when to sow &
-
⏎• High levels of task commitment (motivation)⏎• High levels of creativity⏎EXP: Many gifted students find logical reasoning to be their preferred learning style, & for others linguistics may be (Language Arts & Math are not automatically synonymous –
-
IS, CURE, PREVENTION - home - CONTENTS: Analysis of small paint samples provides the history of paint layers & colors on buildings & can diagnose the cause of paint failures - How to diagnose the
-
Net = hardware (servers, routers, switches, wires, etc.) + the TCP/IP protocol, which makes the data stored on the web servers available to you & me (.. via the URL, entered into your browser-of-choice).⏎Not sure
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 a 3/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or/or
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
contain that have been useful in research.⏎The glands are enclosed in a capsule of and internally divided into . Blood vessels and nerves enter the glands at the and gradually branch
-
465<|endoftext|>“.. modern science seems to have exploded into a multitude of highly specialised areas and distinct disciplines that may at times be interconnected, but that by and large ignore one another. There appears to be an overwhelming trend toward a proliferation of
-
about the progression and flow of my lectures. I believe that students tend to get sort of tired and space out a bit if a lecture on any topic goes on and on in the same way, so I try to work in changes and transitions that
-
be 850, 845, 840, 835, 830, 825 and on and on! On our Number of the Day chart we also look for patterns. We
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
- L13:131 excites 'and' joining parallel phrases
- L4:57 suppresses
- L9:92 suppresses "and" or "or" joining two items
- L1:1250 suppresses
- L10:74 excites
- L0:2146 suppresses
- L16:376 suppresses 'or' joining alternatives
- L0:313 excites "and" joining paired nouns
- L8:70 excites 'and' between stacked superlatives
- L15:78 excites
- L4:1330 suppresses
- L18:868 suppresses 'and' joining two nouns in a pair
- L15:82 suppresses
- L15:114 weakly excites
- L17:2782 weakly excites 'same' in 'at the same time'
- L15:115 weakly excites
- L18:569 weakly excites
- L15:122 weakly suppresses
- L16:99 weakly suppresses
- L2:75 weakly suppresses
- L3:50 weakly excites
- L18:131 weakly suppresses
- L0:114 weakly suppresses
- L15:80 weakly suppresses
Feeds later layers
- L26:1007 strongly suppresses
- L22:198 suppresses
- L20:885 suppresses 'and' joining paired words
- L22:1739 suppresses 'and' starting a place or person name
- L24:205 excites
- L20:2386 suppresses
- L26:2490 suppresses
- L25:1694 suppresses 'and' joining paired names or fields
- L23:2015 weakly suppresses
- L25:1008 weakly suppresses
- L26:347 weakly excites the 're' in self-reliant and similar
- L23:1185 weakly excites
- L25:120 weakly suppresses
- L22:262 weakly excites comma inside big numbers
- L22:791 weakly excites
- L26:513 weakly excites
- L21:173 weakly excites
- L24:124 weakly excites dollar sign before a money amount
- L27:172 weakly excites
- L26:729 weakly excites
- L25:3035 weakly suppresses
- L22:2880 weakly suppresses
- L21:571 weakly suppresses
- L20:1009 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
- 发展
- 技术支持
- 社会
- 和支持
- 发展模式
- 社会发展
- 服务质量
- 社会稳定
- 和个人
- 技术和
Suppresses
- /or
- 煦
- thus
- ROID
- 蔼
- 小编一起
- /OR
- 珅
- rogen
- therefore