"to" between a starting and ending value
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. 7 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
Fires on the word "to" when it sits between a starting point and an end point, like "from 75 to 88 percent" or "from one location to another". Steering did not show a clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
⏎In his<|endoftext|>Refraction is the bending (change in the direction) of a wave (light or sound). When a light ray passes from one transparent medium to another, it changes speed and bends.⏎- Light traveling from air and going into
-
Worldwide, it took an average of 3 decades for the share of 6- to 12-year-old girls in school to increase from 75 to 88 percent. MENA achieved the same increase in 2 decades:
-
1947 and 1948 have increased from 7 percent to 15 percent over those in 1946. That means an
-
moving from one location to another, they are stored in warehouses. At any given time, there are numerous
-
Posted by Marcus on .⏎An ideal monatomic gas initially at 298 K and 1 atm pressure is⏎expanded from a volume of 1 liter to 2 liters. Consider separately⏎isobaric, isothermal and adi
-
iPad to learn and study.⏎And the AP found that the percentage of preschoolers with a social media account increased by almost a third, from about one in ten to about one-half of one percent.⏎Here’s what you need to know about kindergarten
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: something, and I think I should start with the question. Let me see if I can come up with something. Well, perhaps, and maybe, and I'll have to accept. Maybe, and maybe, and perhaps, and maybe,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the older and more dominant bulls around while learning how to be mature elephants.⏎Fathers, just like the juvenile elephant our children follow and learn how to live from us.
-
wastes in our daily life by various ways. We can start from reducing the usage of plastic bags, plastic straws and switch to a compostable
-
. "Sir G. C. Whitlock (During the 1834 Coorg Campaign)". www.archerfamily.org.uk. Archer Family. Archived from the original on 13 November 2013. Retrieved 1
-
:<|endoftext|>Copyright © University of Cambridge. All rights reserved.⏎'Aim High' printed from http://nrich.maths.org/⏎Why do this problem?⏎offers the opportunity
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
- L4:192 excites
- L5:2052 excites
- L2:1147 suppresses
- L0:1660 suppresses
- L5:1922 weakly suppresses 'to' in 'X to Y' phrases
- L4:561 weakly excites
- L3:1119 weakly excites
- L4:1941 weakly suppresses
- L6:1364 weakly suppresses
- L5:640 weakly suppresses
- L5:628 weakly suppresses
- L6:2482 weakly suppresses
- L4:1229 weakly excites
- L6:3050 weakly suppresses
- L2:2379 weakly suppresses
- L1:2826 weakly suppresses
- L0:719 weakly suppresses
- L3:2346 weakly excites
- L3:2601 weakly suppresses
- L2:1110 weakly excites
- L3:1846 weakly suppresses
- L0:1199 weakly suppresses
- L2:1627 weakly suppresses
- L2:2995 weakly excites
Feeds later layers
- L14:706 excites
- L12:2548 excites
- L9:1449 excites
- L12:2721 weakly excites
- L13:2356 weakly excites
- L12:2930 weakly excites "and" joining contrasting pairs
- L8:2415 weakly suppresses words after 'will be' or 'subject to' in formal text
- L11:1861 weakly excites
- L12:2145 weakly suppresses
- L8:3052 weakly suppresses
- L22:818 weakly excites
- L12:925 weakly excites
- L15:584 weakly suppresses
- L24:785 weakly suppresses
- L13:545 weakly excites 'of' or 'to' linking two things
- L18:2880 weakly suppresses
- L17:2642 weakly excites
- L22:1035 weakly suppresses
- L10:708 weakly excites
- L24:2846 weakly suppresses
- L24:761 weakly suppresses
- L12:2208 weakly excites
- L11:1757 weakly suppresses
- L20:74 weakly suppresses abbreviations before a period (Dr, Mr, e.g)
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
- ov
- viron
- obo
- erhalten
- 开放
- 那么多
- 你在
- obl
- 活动
- idable
Suppresses
- icken
- ”⏎⏎
- ätze
- pledges
- horizontally
- Rooney
- 萌
- licit
- findOne
- <header