dictionary and citation boilerplate
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 mixed: they peak on different tokens. 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 'by' and 'for' in copyright and source notes, like 'from WordNet 3.0 Copyright 2006 by Princeton University'. It goes negative on sales and price notices. Turning it up made the model write everyday mixed text, and turning it down made it repeat words.
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.
-
A 8|⏎Article from a library⏎|Web site with no author|⏎The examples above are reproduced from https://owl.english.purdue.edu/owl/resource/747/22/⏎
-
tionary, Creative Commons Attribution/Share-Alike License⏎- n. An underwater explorer.⏎from WordNet 3.0 Copyright 2006 by Princeton University. All rights reserved.⏎- n. an underwater swimmer equipped with
-
avia and Finland with the Kola Peninsula in Russia, traditionally inhabited by the Sami people.⏎from WordNet 3.0 Copyright 2006 by Princeton University
-
which as explained here is only thought to be a possible explanation for dark matter in combination with Cold Dark Matter theory and not on its own.⏎This page was last updated June 27, 2015.<|endoftext|>05-1
-
, thin, or emaciated appearance; drawn; said of the face or the expression.⏎from WordNet 3.0 Copyright 2006 by Princeton University. All rights reserved.⏎- adj. somewhat ill or prone to illness
-
. FormsCan be used for anything<|endoftext|>We found 1,708 reviewed resources for state of the union address⏎How can word choice affect a political speech? Middle
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: 2, 25, 10, 14, 26, 11, 19, 12, 10, 10, 10, 10,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
6⏎- Saver Delivery: Yes⏎- 1st Class Delivery: Yes⏎- Courier Delivery: Yes⏎- Store Delivery: Yes⏎Prices are for internet purchases only. Prices and availability in WHSmith Stores may vary significantly⏎©
-
and directed by A. J. Hildebrand.)⏎Open content licensed under CC BY-NC-SA⏎Given an integer , the exponential sum version of the Gauss sum modulo is defined by⏎and the quadratic residue version is defined by⏎where is
-
our deepest concerns, judges us and guides us through our lives. A traditional Jewish concept of God is one that is interactive and intimately personal.⏎This week's Torah portion, Vaerah, begins with God hearing, and ultimately reacting to, the Israel
-
Saturday<|endoftext|>15.1: The Science Fair (5 minutes)⏎The purpose of this warm-up is for students to reason about two objects that have the same volume but different surface
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
- L2:299 weakly excites
- L6:2939 weakly excites
- L4:182 weakly suppresses
- L8:774 weakly excites
- L5:706 weakly excites the number 1 starting a list
- L3:2117 weakly excites
- L5:1912 weakly excites
- L5:1588 weakly excites
- L7:3025 weakly suppresses
- L3:984 weakly excites "to" after a noun saying what to do
- L8:1221 weakly suppresses
- L0:2954 weakly suppresses
- L8:1474 weakly excites
- L0:413 weakly suppresses
- L5:164 weakly excites
- L8:224 weakly suppresses
- L2:1437 weakly suppresses
- L3:2915 weakly excites
- L3:1604 weakly excites
- L1:2967 weakly suppresses
- L1:1431 weakly excites
- L7:1438 weakly suppresses
- L1:2658 weakly suppresses
- L4:2256 weakly suppresses
Feeds later layers
- L10:1836 suppresses
- L17:1779 suppresses
- L14:1864 suppresses
- L14:1850 suppresses
- L17:2473 suppresses
- L18:1401 weakly suppresses
- L13:1162 weakly suppresses
- L13:2401 weakly suppresses
- L12:1378 weakly suppresses
- L22:1051 weakly suppresses
- L15:1909 weakly suppresses blank slot before a number
- L15:368 weakly suppresses
- L19:2743 weakly suppresses
- L14:1697 weakly suppresses
- L16:2069 weakly excites
- L15:1657 weakly suppresses
- L10:3023 weakly suppresses
- L13:2449 weakly suppresses
- L12:2168 weakly suppresses
- L15:2315 weakly suppresses
- L12:2811 weakly suppresses
- L23:1857 weakly suppresses
- L17:1488 weakly suppresses
- L16:3054 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
- 骅
- овый
- HandlerContext
- aca
- 的数据
- ólogo
- oci
- andan
- opport
- roup
Suppresses
- 저
- (nameof
- rounds
- (round
- 凌晨
- round
- ension
- 关门
- 一向
- ltd