'reasons' or 'permitting' after a qualifier
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. 1 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 that finishes a short phrase giving a justification, such as 'for health reasons', 'weather permitting' or 'if time allows'. When turned up, the model's writing did not show a clear link to this idea.
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.
-
" is commonly used for distribution systems with voltages above 1 kV and generally applied up to and including 52 kV . For technical and economic reasons, the service voltage of medium voltage distribution networks rarely exceeds 35 kV
-
world around them, but that they can make choices that reduce their impacts on the land, water, air, and other living and/or natural systems. If time allows, ask the students to write about some of the connections that come to mind in
-
an exception to copyright protection called fair use. This important part of copyright law supersedes the author's exclusive right to duplicate and distribute his works for public policy reasons, recognizing the tension between an author's ownership rights and society's desire to use
-
(EGR) systems are evaluated, as such systems will probably be applied on future engines due to expected new NOx emission regulations, as well as for economical reasons of the engine production.⏎To confirm the potential, a reduced-air-flow test has
-
an individual and physical examination must be performed in a respectful and culturally sensitive manner. Privacy is essential and the nurse must be aware of their own ...⏎You
-
D.C., at age 67.⏎In 1930, the chief justice of the United States, William Howard Taft, resigned for health reasons. (He died just over a month later.)⏎In 1943
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: "I need to take more time to get a good look, but also to take care about the time I have spent on my work. I will take a look at the photos of the best fashion brands, and try to find the right
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
& Aid⏎- Economy & Trade⏎- Human Rights⏎- Global Governance⏎- Civil Society⏎Sunday, September 19, 2021⏎MONTE
-
asters,”UN News Service, 13 June 2006.<|endoftext|>On Thursday, December 15, 2016 the Civil Society Working Group on Synthetic Biology circulated a short fact sheet to help clear up four of
-
who have studied climate change: Human activities since the Industrial Revolution have been pumping greenhouse gases into the atmosphere, and this is not good.⏎"Warming of the climate system is unequivocal," stated the leading authority on climate change, the Intergovernment
-
-forms on it evolving slowly over hundreds of millions of years.⏎That seems to contradict the biblical version of creation, with God making the earth and its many life-forms rapidly.and probably not all that long ago.⏎I brought my boys Austin and
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
- L0:1224 suppresses
- L0:178 suppresses
- L0:1301 suppresses
- L0:2067 weakly suppresses
- L0:924 weakly suppresses
- L0:921 weakly excites
- L0:1571 weakly excites
- L0:1369 weakly suppresses
- L0:794 weakly excites
- L0:1834 weakly suppresses
- L0:1299 weakly suppresses
- L0:1799 weakly excites
- L0:225 weakly suppresses
- L0:1118 weakly suppresses
- L0:909 weakly excites
- L0:1203 weakly excites
- L0:2591 weakly excites
- L0:1582 weakly suppresses
- L0:2229 weakly excites
- L0:659 weakly suppresses
- L0:176 weakly excites
- L0:1213 weakly suppresses
- L0:837 weakly suppresses
- L0:1674 weakly excites
Feeds later layers
- L2:1254 excites
- L9:1655 weakly suppresses
- L15:2193 weakly excites
- L9:199 weakly excites
- L21:1004 weakly excites
- L2:2494 weakly excites
- L11:604 weakly excites
- L10:783 weakly suppresses
- L16:1604 weakly excites
- L13:2951 weakly excites
- L11:558 weakly suppresses
- L4:1387 weakly excites
- L10:1040 weakly excites
- L6:413 weakly suppresses
- L21:258 weakly excites
- L15:2203 weakly suppresses
- L10:2317 weakly suppresses
- L24:569 weakly suppresses
- L5:2634 weakly suppresses
- L8:3037 weakly suppresses
- L5:767 weakly suppresses
- L15:296 weakly excites
- L8:2264 weakly suppresses
- L12:1147 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
- stood
- Jens
- skl
- ần
- ijing
- Devin
- (so
- mad
- Subview
- (Edit
Suppresses
- ste
- omencl
- Caval
- lyph
- 走路
- 复古
- ughters
- 当时的
- itudes
- 巡回