"for" in organisation names
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. 23 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 "for" inside the names of organisations, like "National Institute for Play", "International Union for Conservation" or "Agency for Healthcare Research". When turned down, the model got stuck repeating the word "control", so that is the only steering hint and it's weak.
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.
-
), the Federal Institute for Drugs and Medical Devices (BfArM) and IFA. 0
-
his thoughts on the meaning of happiness, and how he understands spirituality as "contemplative science."⏎Stuart Brown, a physician and director of the National Institute for Play, says that pleasurable, purposeless activity prevents violence and promotes trust,
-
61 and 1964 by biologists in Russia. Hence, it is also called as the Russian Red Data Book.⏎Currently, the International Union for Conservation of Nature maintains the Red Data Book. IUCN is the world’s
-
nforschung (HIH)⏎CRISPR meets single-cell sequencing in new screening method⏎19.01.2017 | CeMM Forschungszentrum für Molekulare Medizin der Österreichischen Ak
-
fruit bat (Styloctenium mindorensis) is found only on the Philippine island of Mindoro. Courtesy of ASU's International Institute for Species Exploration.⏎To date some 1.8 million species have been described,
-
; the study results are based on the widely accepted Greenhouse Gas Protocol [GHGP] established by the World Resources Institute [WRI] and World Business Council for Sustainable Development [WBCSD and were announced during an AMD sponsored media panel on
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: What would control the control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control control
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
microbiomes or dysbiosis and increased intestinal permeability as well as other changes. Gut microbiota may contribute to MS pathogenesis by modulating host immunity through the regulation of multiple metabolic pathways.⏎The “Gut Feeling”: Breaking Down the Role
-
omyces cerevisiae, are well studied and some of the most important food production micro-organisms. S. cervisiae has been used for centuries for the production of wine and beer as well as bread. But other species, the non
-
are just going to explain how hemp differs from marijuana.⏎Hemp has a rich history. It has been grown for many millennia for various purposes, for example, the production of textiles, building materials, food, and skin and hair care products.
-
intment, compresses,or alcoholic digestion) for treatment of disorders of the locomotor system and gastrointestinal tract. The leaves and stems have also been used for the treatment of the same disorders
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
- L20:585 excites
- L20:755 excites
- L16:2182 weakly excites
- L17:472 weakly excites
- L2:2048 weakly suppresses
- L19:768 weakly suppresses
- L10:1638 weakly excites
- L19:769 weakly suppresses
- L20:2262 weakly excites
- L21:2891 weakly suppresses
- L1:386 weakly excites
- L19:1608 weakly suppresses
- L19:2371 weakly suppresses
- L1:2905 weakly excites
- L7:92 weakly excites
- L9:200 weakly suppresses
- L17:1392 weakly suppresses
- L17:1488 weakly suppresses
- L20:2543 weakly suppresses
- L19:655 weakly suppresses
- L17:960 weakly excites
- L21:239 weakly suppresses
- L19:752 weakly suppresses
- L11:895 weakly suppresses
Feeds later layers
- L27:612 suppresses
- L26:346 suppresses
- L24:480 suppresses
- L27:478 suppresses
- L27:1199 suppresses
- L24:2515 weakly excites
- L24:173 weakly excites
- L27:1092 weakly suppresses
- L24:747 weakly suppresses
- L25:868 weakly excites
- L27:666 weakly excites
- L24:2218 weakly excites
- L25:1353 weakly suppresses "in" before "good" condition
- L24:2053 weakly excites
- L25:2663 weakly excites
- L27:2088 weakly suppresses
- L25:793 weakly suppresses
- L25:1922 weakly excites
- L24:738 weakly suppresses
- L25:1512 weakly excites
- L26:501 weakly suppresses
- L27:1560 weakly suppresses
- L27:208 weakly suppresses
- L25:443 weakly suppresses
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
- ulas
- ضا
- /Getty
- Comey
- <Task
- ți
- образ
- slideUp
- orem
- 试验区
Suppresses
- control
- etermination
- evaluation
- etection
- -control
- construction
- Control
- 处理
- evaluation
- 控制