'for' giving a reason or cause
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. 9 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
The neuron fires on the word 'for' when it gives a reason, like 'killed just for fun', 'sent off for headbutting', or 'jail for a conviction'. Its negative side is weak and mixed. Turning it up did not produce text about reasons or causes.
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.
-
stand to see animals killed just for fun. Her sense of injustice makes her angry and through her anger she develops a
-
that took place at Wounded Knee, Mr. Camp was the only one to serve time. He spent two years in prison in Leavenworth, Kan., for assaulting a postal inspector, a charge that Camp-Horinek disputes.⏎In
-
May to July 2006, and towards the end of that period France lost the World Cup final after Zinedine Zidane was sent off for headbutting an Italian player. Despite the national defeat,<|endoftext|>By Connie Vit
-
The Magic Finger.⏎I can't stand hunting. I just can't stand it. It doesn't seem right to me that men and boys should kill animals just for the fun they get out of it.⏎The Girl 1⏎The Magic Finger
-
earlier admission that the Church made a⏎grievous error when it persecuted Galileo for arguing that the⏎Earth revolved around the sun. The article further described the
-
the last 25 years or so."⏎Powell interviewed Tupac a number of times for Vibe magazine, including while the rapper was serving time in jail for a sexual abuse conviction. He joined NPR's Renee Montagne to discuss Tup
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 am the teacher and I am tired of this. I am the teacher and I am tired of this. I am the teacher and I am tired of this. I am the teacher and I am tired of this. I am the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
une Function, 2017, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5707683/#:~:text=It%20is%20a%20potent%
-
everyday terms, it's what is needed to move things, heat them... More into the light bulb⏎- The excess energy returns to the breaker panel through a neutral wire, completing a circuit.⏎This process continues until you turn off the light bulb
-
Argentina.⏎Historically, lithium has been mined either in open pits or underground or extracted from brine using large, plastic-lined solar evaporation ponds. Producing lithium from evaporation ponds can take months and has drawn criticism from some environmental groups for
-
and then its right child.⏎T / \ I S / \ D E<|endoftext|>Original URL: http://www.theregister.co.uk/2006/08/21/mars_geysers/⏎Martian pole freck
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
- L5:1090 excites
- L6:2553 excites 'for' after praise words like 'known for'
- L6:1943 weakly excites
- L3:733 weakly excites
- L8:2745 weakly suppresses
- L6:2143 weakly excites
- L6:1671 weakly excites
- L6:2941 weakly suppresses
- L4:1243 weakly suppresses
- L5:60 weakly excites
- L5:1285 weakly suppresses
- L0:1543 weakly excites
- L3:2019 weakly suppresses
- L3:381 weakly excites 'for' in research institute names
- L7:3049 weakly suppresses
- L5:216 weakly suppresses
- L4:2502 weakly suppresses
- L3:2322 weakly suppresses
- L1:784 weakly excites
- L5:110 weakly suppresses
- L3:163 weakly suppresses
- L0:1356 weakly excites
- L4:1395 weakly excites
- L5:1213 weakly suppresses
Feeds later layers
- L10:1855 excites
- L13:2509 excites
- L24:2492 weakly excites
- L23:331 weakly excites
- L22:361 weakly excites
- L10:1054 weakly excites 'from' after movement or transfer verbs
- L17:2931 weakly excites
- L10:2075 weakly excites
- L15:1642 weakly excites
- L14:965 weakly excites
- L12:2694 weakly excites unclear: 'purpose' and words near it
- L18:2859 weakly excites 'for' after an award or honour
- L20:56 weakly excites
- L14:2288 weakly suppresses
- L18:2238 weakly suppresses
- L15:1273 weakly excites
- L24:2885 weakly suppresses
- L11:821 weakly excites
- L15:765 weakly suppresses
- L10:2311 weakly excites indefinite article before a noun
- L15:1933 weakly excites
- L10:871 weakly excites
- L26:1081 weakly excites
- L21:2337 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
- 引起的
- asp
- avigate
- 敌
- Dirty
- avig
- _));⏎
- due
- PostBack
- legitimate
Suppresses
- ورة
- imators
- acements
- 不多
- invers
- ワーク
- .icons
- icus
- essor
- anguages