"if not" in "X, if not Y" phrases
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. 2 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 "not" when it comes right after "if" in phrases like "if not all" or "if not millions". When it was turned up, the model's writing drifted into odd, loose text, with no clear shared 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.
-
tanks, large flags, chest thumping marches, discipline-for-itself and martyr walls help inspire nationalism and patriotism? This becomes a particularly good question, if not a surprising one, because, historically speaking, ideas of nationalism and patriotism have never
-
nothing until it is achieved, and with the initial focus often projected at how to pass the entry exams, the aftermath of admission has been a nightmare for thousands if not millions of students.⏎English is one the most widely spoken international languages and many would
-
is one of the most important, if not the most important, metrics in valuing the bond.⏎Now, let's look
-
as cabinet minister Michael Chong recognized, is a recognition of "ethnic nationalism".⏎In the U.S., we have a number of ethnic nations, most, if not all, being Amerindian in nature. But, citizens of the Navajo
-
high-presure steam riding rocket'. In the early part of the 19th century, steam propulsion was regarded by some as something of an unreliable, if not downright dangerous, fad, with the risk of boiler explosions particularly to be feared
-
eating plan.⏎Beyond promoting every nutrition expert’s mantra of “eat more vegetables and fruits”, the standard raw food diet is true to its name – most, if not all, foods need to be “raw”. This means dieters are prohibited from
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: that I have a lot of time to be, and to do so to do so to do so to do so to do so to do so to do so to do so to do so to do so to do so to do so to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Commission (EEOC) authority to sue in federal courts when it finds reasonable cause to believe that
-
of modern physics--the empirical basis of quantum mechanics, wave-particle duality and the uncertainty principle, and the action-at-a-distance theory of Wheeler and Feynman. 1961 ed.|⏎|Keywords||Physics History Physics
-
, when he was a postdoctoral fellow. (This was his first return visit since then.) He had just gotten his PhD at Caltech under Richard Feynman, and was busy trying to understand the “zoo” of hadronic
-
’m a fan of physicist Richard Feynman, and one of his most memorable quotes explains that “Science is the belief
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
- L1:189 weakly suppresses
- L1:1581 weakly excites
- L0:492 weakly excites
- L1:367 weakly suppresses
- L0:178 weakly suppresses
- L0:165 weakly excites
- L1:1464 weakly excites
- L1:2930 weakly excites
- L0:1224 weakly suppresses
- L0:1301 weakly suppresses
- L1:1923 weakly excites
- L1:1724 weakly excites
- L1:673 weakly excites
- L0:778 weakly suppresses
- L0:2268 weakly suppresses
- L0:1 weakly suppresses
- L0:312 weakly suppresses
- L0:3000 weakly suppresses
- L1:331 weakly suppresses
- L1:2143 weakly suppresses
- L1:951 weakly excites
- L0:2167 weakly suppresses
- L0:1075 weakly excites
- L0:2914 weakly suppresses
Feeds later layers
- L11:1519 weakly suppresses
- L9:1892 weakly excites
- L4:265 weakly excites
- L10:2223 weakly excites
- L22:136 weakly excites
- L19:2886 weakly excites
- L11:1975 weakly excites
- L20:969 weakly excites
- L4:1873 weakly suppresses
- L10:1055 weakly suppresses
- L22:1749 weakly suppresses
- L9:929 weakly suppresses 'each' in 'each other'
- L7:2499 weakly excites
- L15:1929 weakly suppresses
- L15:142 weakly suppresses
- L7:1921 weakly excites
- L14:2659 weakly excites
- L10:1499 weakly excites
- L23:2556 weakly suppresses
- L21:1748 weakly excites
- L16:621 weakly excites
- L6:2531 weakly excites
- L13:364 weakly excites
- L22:2705 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
- ugo
- ìm
- enge
- ilm
- �
- 钢筋
- ung
- 实战
- mode
- eng
Suppresses
- CallCheck
- eness
- antly
- bons
- egra
- worthy
- panion
- ibly
- criptions
- ativity