'rem' at the start of words
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. 26 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 syllable 'rem' at the start of a word, like remodeled, remembrance, remnant and remission. When turned up, the model got stuck repeating a single word, so the steering shows nothing clear about meaning.
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.
-
destroyed by fire in October 1863.⏎For the next 30 years the court moved into a number of temporary quarters: a school, a remodeled jail, a two-story building on the west side of the square and a
-
the first fifteen verses of Psalms 105:1-45. This is a general ascription of praise which merges into a call to remembrance of the works of God, and of His government covenant with the people.
-
, White Dwarfs or Pulsars?⏎A personal account by Jocelyn Bell Burnell on the discovery of pulsars.⏎Article from Wikipedia on this remnant that can result from the gravitational collapse of a massive star.⏎Overview of what
-
1 poster designed by innovative graphic artist Nick Hayes, which served as a step-by-step guide to researching potential Passchendaele graves and lots of inspiration for remembrance activities within groups and local community, including wreath making and playing the
-
/proclivities, etc. Psychometric profiles of civilian<|endoftext|>Recently we were called out to investigate a remnant shipwreck that had materialized on the shoreline near Georgetown, Prince Edward
-
law purged with blood, and without shedding of blood is no remission” (Hebrews 9:22). To purge is to
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: my love for plumskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskinskins
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
lives on earth.⏎Microbiology is the study of the living organisms which are small and cannot be seen with the naked eye. Some of the microorganisms include viruses, fungi, bacteria, protozoa and algae. Microbiology helps in
-
point (2, 4)?⏎November 18, 2010⏎biology / plasmolysis of human red blood cells⏎Plasmolysis of human red blood cells would occur
-
ers ampicillin resistance to bacterial hosts.⏎- Neo gene from Tn5, which confers antibiotic resistance to kanamycin in bacteria and to geneticin in eukaryote cells.⏎- Mutant FabI gene (mFabI
-
—and tasty!—way to make the parts of the cell more concrete for students is to create edible cells. Using materials such as Jell-O® (to represent cytoplasm) and sprinkles (to represent ribosomes) will not only
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
- L22:2646 excites
- L24:105 weakly suppresses
- L18:541 weakly suppresses
- L22:1640 weakly excites
- L17:253 weakly suppresses
- L3:1240 weakly suppresses
- L16:3008 weakly excites
- L22:2518 weakly suppresses
- L0:1849 weakly excites
- L16:60 weakly suppresses
- L6:2904 weakly suppresses
- L0:1365 weakly suppresses
- L24:2908 weakly suppresses
- L20:1852 weakly suppresses
- L3:1325 weakly excites
- L6:2953 weakly excites
- L8:3018 weakly excites
- L5:2060 weakly excites
- L9:31 weakly suppresses
- L1:2432 weakly suppresses
- L6:1759 weakly suppresses
- L10:612 weakly suppresses full stop ending boilerplate or a title
- L20:1886 weakly suppresses
- L16:44 weakly excites
Feeds later layers
- L27:2842 weakly excites
- L27:365 weakly excites
- L27:2939 weakly suppresses
- L27:1322 weakly excites
- L27:492 weakly excites
- L27:1662 weakly excites
- L27:2390 weakly excites
- L27:2331 weakly excites
- L27:1541 weakly suppresses
- L27:2376 weakly suppresses capital D starting a split word
- L27:155 weakly excites
- L27:2992 weakly suppresses
- L27:2577 weakly excites
- L27:782 weakly excites
- L27:2809 weakly suppresses
- L27:976 weakly excites
- L27:1573 weakly excites
- L27:1754 weakly excites
- L27:3055 weakly suppresses lowercase 'l' starting a word
- L27:489 weakly excites
- L27:1770 weakly suppresses
- L27:2885 weakly excites
- L27:2528 weakly excites
- L27:496 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
- 轵
- men
- ijkl
- dbe
- rze
- физ
- Shutterstock
- _cmos
- ukt
Suppresses
- .Extension
- ương
- trä
- perimental
- conserv
- 즈
- 袍
- angs
- まと
- kins