'by' or operator after a number in arithmetic
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. 11 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 tends to fire on the word just before a maths step, especially 'by' in instructions like 'multiply the number of stairs by' or 'divide the total weight by'. When turned up, one answer slipped into multiplying numbers, which fits, but most of the text was unrelated.
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 comment if you LOVE your math series! I'd love to check it out!⏎I like the tricks (9s trick, Doubles, Doubles Plus 1) that Saxon uses to teach facts, but even after learning and practicing
-
- Determine the height of the landing from the bottom of the stairs. this will⏎- Decide the number of stairs and their width. multiply the number of stairs by .28 to get the total landing space required⏎1. Cascading deck
-
TMJ disorder develops, signs and symptoms to watch out for, and tmj treatment Albuquerque options to address the condition effectively.⏎What is TMJ?⏎Temporomandibular joint disorder, or TMJD, is a condition that affects a
-
s–1 ( Eddington limit for neutron stars, ), and then typically decaying in a few tens of seconds3.⏎Multiplying the burst luminosity by its duration we get an estimate of the total burst energy 1039
-
illin . Tell your doctor or pharmacist if you have any medical conditions, especially if any of the following apply to you:⏎if you are pregnant, planning to become pregnant, or are breast-feeding⏎if you are taking any prescription or nonpres
-
.⏎- Add an additional 25 percent of the total weight to allow a margin of error.⏎- Divide the total weight including the additional 25 percent by the number of wheels supporting the cart. The resulting
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: the most common, most accurate, and even, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
fairly vanilla neural networks called Alice, Bob, and Eve.⏎Ref: Learning to Protect Communications with Adversarial Neural Cryptography. arXiv - Computer Science > Cryptography and Security (21 October 2016) | ar
-
leader for fourteen years.⏎Subjects: History — Social Sciences.<|endoftext|>Newegg.com - A great place to buy computers, computer
-
, his own understanding of them.⏎Subjects: Judaism and Jewish Studies.<|endoftext|>Did you ever wonder the meaning of the word Christmas
-
kworm-BEVS is a promising alternative strategy to produce the functional and thermostable reverse transcriptase.⏎All Science Journal Classification (ASJC) codes⏎- Insect Science<|endoftext|>An account of the religion of
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
- L4:303 suppresses
- L9:679 excites
- L3:164 weakly suppresses
- L7:1756 weakly suppresses
- L8:746 weakly suppresses
- L6:1674 weakly excites
- L9:1206 weakly suppresses
- L10:1110 weakly suppresses
- L10:891 weakly suppresses
- L9:327 weakly excites
- L3:348 weakly suppresses
- L2:862 weakly excites
- L6:1610 weakly excites
- L8:881 weakly suppresses
- L6:158 weakly excites
- L8:1083 weakly excites "x" or "by" between two dimensions
- L4:2574 weakly excites
- L1:2926 weakly suppresses
- L10:1285 weakly suppresses
- L4:378 weakly excites
- L10:1331 weakly suppresses
- L5:2771 weakly excites
- L10:2607 weakly suppresses
- L9:2334 weakly excites
Feeds later layers
- L14:1654 weakly excites
- L16:2196 weakly excites
- L19:2698 weakly excites
- L17:2322 weakly excites
- L18:1637 weakly excites
- L16:1818 weakly excites "two" and the link between two things
- L14:1082 weakly suppresses
- L16:490 weakly excites
- L17:3010 weakly excites
- L23:2547 weakly excites
- L20:2005 weakly suppresses
- L17:1321 weakly excites
- L17:1036 weakly excites
- L18:1887 weakly excites
- L15:1078 weakly suppresses
- L20:1001 weakly excites
- L16:2323 weakly suppresses
- L16:2856 weakly suppresses
- L12:944 weakly suppresses
- L19:2323 weakly suppresses
- L17:462 weakly suppresses
- L12:1978 weakly suppresses
- L18:478 weakly suppresses
- L18:685 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
- `
- omb
- ~⏎⏎
- 印
- (:,:,
- `(
- perceive
- Providence
- onium
- absorb
Suppresses
- olarity
- ẩm
- marks
- rier
- ực
- ihan
- ramento
- athing
- asher
- lean