"only" in "it is only" constructions
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. 17 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 word "only", especially in phrases like "it is only due to" or "but only if", where the writer limits or qualifies a claim. Turning it up did not give a clear theme in the model's replies, so its effect on writing is unclear.
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.
-
deep cleaning and/or scaling and root planing for more severe cases<|endoftext|>Considering the fact that half of you reading this right now are probably wearing one, It’s only right to start this week off with the most commonplace of all acronymic items–
-
cars are coming - if you can fill up⏎LONDON (Reuters) - Britain's hydrogen fuel cell car fleet may hit top gear within five years, but only if there is enough investment in filling stations, the UK Hydrogen and Fuel Cells
-
finance and law. As the world changes, most of our financial matters have moved seamlessly online. Our digital identity is lagging in terms of adoption but it is only due to the sensitivity of the matter at hand.⏎Digital identity is used in relation
-
kill very quickly. And it is not the only danger that we may face at home and in public places. A few basic precautions can save your life — but only if you follow them!⏎Fire safety at home⏎Your biggest lifes
-
instinctive manner: it is to know the reasons behind these rules and, not less importantly, to know their limitations.⏎Regrettably, it seems it is only in times of trouble and uncertainty when we begin to realize tradition is a living and
-
looking ahead to a problem on the horizon: Many Covid-19 survivors will have long-term lung injuries, and medicine has little to offer them.⏎There’s only scant data on how many patients who recover from Covid-19 are left with
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 been, for all the time, but have been, for all the time, but have, for all the same time, I have been, in the same time, the same time, but I have been, in all the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
01).⏎Predator conservation has to date operated primarily within the boundaries of existing protected areas, but there are several limitations in relying solely upon this approach: only around 5% of the world’s land area is formally protected (Gittleman
-
differences they noticed and then embed this in the class website".<|endoftext|>Instructions: In this discussion, you will need to post your response (no fewer than 200 words and including citation and research) before you can see a
-
an activity that has many benefits for the body⏎Yoga is more than just physical activity. In the Indian tradition, it is a real way of life comprising
-
/stereo/article/view/6807<|endoftext|>The term “education technology” (or “edtech,” as it’s commonly known) refers to more than just learning via computer. Edtech has existed in various forms for as long as education
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
- L4:661 strongly excites
- L10:652 suppresses
- L9:2343 suppresses
- L4:2502 suppresses
- L2:2153 suppresses
- L4:3039 excites
- L8:140 excites
- L15:2689 excites the word 'only' mid-sentence
- L11:1697 weakly excites
- L14:1400 weakly suppresses
- L15:1139 weakly suppresses
- L15:2637 weakly suppresses
- L13:948 weakly suppresses
- L11:678 weakly suppresses
- L14:536 weakly suppresses
- L7:63 weakly excites
- L15:469 weakly excites
- L1:2119 weakly excites
- L7:2753 weakly excites
- L16:1422 weakly suppresses
- L3:2933 weakly suppresses
- L11:2367 weakly suppresses
- L4:1830 weakly excites
- L13:2845 weakly excites
Feeds later layers
- L23:2337 excites
- L18:1130 excites
- L19:2011 excites
- L21:2533 excites
- L18:2478 excites
- L18:2883 excites
- L21:1226 suppresses
- L20:1279 excites
- L18:590 excites
- L18:2755 weakly suppresses
- L20:2305 weakly excites
- L19:1298 weakly excites
- L18:2181 weakly suppresses
- L26:271 weakly suppresses
- L21:1349 weakly excites
- L20:2339 weakly excites
- L23:2083 weakly excites 'a' meaning one single
- L20:2494 weakly suppresses
- L23:1498 weakly excites
- L27:2346 weakly excites
- L23:2170 weakly excites
- L18:2206 weakly suppresses
- L25:549 weakly suppresses
- L24:242 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
- izioni
- ertiary
- iciones
- 一件事
- 砜
- sofar
- "..
- ões
- motivate
- oser
Suppresses
- all
- 歹
- -cr
- ibernate
- 所有的
- AY
- 年代
- 任意
- ursal
- ay