capital E starting an acronym
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 responds to a capital E that opens an acronym in brackets, like the E in '(EIGA)' or '(ECD)'. When it is pushed negative it tends to respond to a lowercase e starting a word fragment instead. Steering outputs were not clearly connected to this pattern.
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.
-
tissue.⏎- Pull the<|endoftext|>Listen to radar echoes from satellites and meteors, live on listener-supported Space Weather Radio.⏎| || |⏎ORIONID METEOR UPDATE: Earth is passing through a broad stream of debris from Halley's
-
1; G-1.2; G-1.6; G-1.9; G1.10, European Industrial Gases Association (EIGA) – IGC Doc 123/1<|endoftext|>How Canadian
-
researchers and regular people can use to do science and educate.<|endoftext|>The learning program at Rona Glynn Preschool is designed using the Early Years Learning Framework (EYLFv2.0) Australia’s national Early Years Framework used cohes
-
combined with Scavenge Air Moisturising (SAM) or Exhaust Gas Recirculation (EGR) systems are evaluated, as such systems will probably be applied on future engines
-
to assist you during this process.⏎State and District Resources⏎Comprehensive School Reform: Five Lessons from the Field -- Available from the Education Commission of the States (ECS) Education
-
food was haraam, his drink was haraam, his clothing was haraam,<|endoftext|>You may have passed your International English Language Testing System (EILTS), TOEFL, TOEIC, a college admission essay
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: When people keeppickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingpickingp
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
patient care," Venneti says. The study is published in Science Translational Medicine.⏎Traditional prognostic markers based on tumor grade have proven unreliable for ependymomas. Understanding a patient's prognosis can help doctors make treatment recommendations,
-
) is the UK government agency that helps provide this capability.⏎Satellite data is essential for comparing to computer simulation models, historical (paleo) data from e.g. ice cores and tree rings, and to test and improve our scientific understanding
-
has been associated largely with human intervention on coastal processes in the form of industrialization, recreation, watershed runoff, agricultural land uses, dredge and fill operations, eutrophication, and unsustainable fishing practices;⏎Recalling the catastrophic consequences to fisheries
-
terrestrial discharge, atmospheric inputs, mixing, and ecological interactions. As that process unfolds, interim models representing the best understanding to date are applied to real world problems of eutrophication and water quality created by human habitation of the
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
- L18:221 excites
- L22:2428 excites
- L17:542 suppresses
- L16:56 suppresses
- L21:361 suppresses
- L17:594 suppresses
- L2:1749 excites
- L19:598 excites
- L20:374 suppresses
- L24:586 excites
- L20:1398 excites
- L0:2901 excites
- L10:2401 excites
- L16:118 suppresses
- L0:54 excites
- L17:366 suppresses
- L14:392 weakly suppresses
- L16:205 weakly excites
- L1:141 weakly suppresses
- L18:1104 weakly excites
- L19:645 weakly suppresses
- L2:1795 weakly excites
- L18:139 weakly excites
- L0:1242 weakly suppresses
Feeds later layers
- L27:2255 weakly suppresses
- L27:2880 weakly suppresses
- L27:736 weakly suppresses
- L27:2763 weakly excites
- L27:1436 weakly suppresses
- L27:2963 weakly suppresses the 'st' piece inside words
- L27:758 weakly suppresses
- L27:1427 weakly suppresses
- L27:1810 weakly suppresses
- L27:1307 weakly excites
- L27:1849 weakly suppresses
- L27:2188 weakly suppresses
- L27:1147 weakly suppresses
- L27:1469 weakly excites
- L27:1802 weakly suppresses
- L27:2389 weakly suppresses
- L27:2597 weakly suppresses
- L27:2417 weakly excites
- L27:2452 weakly excites
- L27:1773 weakly suppresses
- L27:2909 weakly excites
- L27:2662 weakly suppresses
- L27:1476 weakly suppresses
- L27:792 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
- 就好像
- ièrement
- ['',
- "'",
- .pem
- ục
- i
- 涪
- .fd
- unte
Suppresses
- mö
- 翰
- 全国人民
- mer
- finite
- p
- m
- nable
- motion
- Gen