people nouns after a modifier
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 mixed: they peak on different tokens. 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 words for groups of people, such as 'people', 'patients', 'parents' and 'children', especially when a word like 'closely', 'less time' or 'each year' comes just before them. Turning it up did not produce a clear, related change in the model's writing.
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.
-
for Alzheimer's and dementia.⏎Heidi Wengreen, RD, PhD, Assistant Professor of Nutrition at Utah State University, and colleagues examined associations between how closely people adhered to the DASH diet and risk of cognitive decline and dementia among older
-
how long they will be sick, exposed to dialysis or ventilators. The less time patients spend in there the better off they are.”⏎For patients who spend several weeks in
-
established.<|endoftext|>The MOST Theological Collection: Mary in Our Life⏎"Questions for Discussion"⏎For private, class
-
an individual school and on a system-wide basis.⏎Some of the challenges to working with parents:⏎1. By nature, they are a moving target: each year parents enter and depart the school system as they sequence from elementary, to middle, to
-
barriers brought partly by COVID-19, the education system and its governing bodies can offer strategic lessons and good practices in almost any situation. Wherever education is disrupted for a protracted period, the three lessons mentioned below can actively ensure
-
traditions. Compliance with tribal norms held people together strongly, and enabled the tribe to deal effectively with external threats. If you
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: the word "a" in the word "b" in the word "c" The word "d" The word "s" and the word "d" The word "s" and the word "d" The word "s
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
able to share that with their children gave parents a feeling of normalcy, and made some feel more in control of the situation they were in.⏎“As health professionals, we
-
do the homework and don’t ask questions, or engage in discussion. Instead, they just watch or listen and let knowledge wash over them. The challenge of teaching passive learners is especially apparent for kindergarteners, who have not yet developed independent
-
6 Examination.⏎We had mentioned some tips for cracking the HSC Rajasthan Board exam here: HSC Study Tips to
-
patch to generate a faithful copy of the other.⏎Unfortunately, some changes to directory trees cannot be expressed using⏎current patch formats; also,⏎patch does not handle some of the⏎existing formats. These shortcomings motivate the following suggested⏎patch do not
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
- L1:254 excites
- L1:386 excites
- L1:201 suppresses
- L1:1874 suppresses
- L1:266 weakly excites
- L1:958 weakly suppresses
- L1:1752 weakly suppresses
- L1:535 weakly excites
- L1:1055 weakly excites
- L1:1495 weakly excites
- L1:2352 weakly suppresses
- L1:384 weakly excites
- L1:1231 weakly suppresses
- L1:848 weakly suppresses
- L0:2129 weakly excites
- L1:969 weakly excites
- L1:861 weakly excites
- L1:127 weakly suppresses
- L0:2616 weakly suppresses
- L1:302 weakly excites
- L1:632 weakly excites
- L1:456 weakly suppresses
- L1:1870 weakly suppresses
- L1:807 weakly excites
Feeds later layers
- L3:472 excites
- L4:250 excites
- L3:195 excites
- L3:352 excites
- L3:126 excites
- L3:997 excites
- L7:310 suppresses
- L3:84 suppresses
- L15:2616 suppresses
- L8:917 excites
- L3:426 suppresses
- L4:286 excites
- L3:226 weakly suppresses
- L12:710 weakly excites
- L6:738 weakly excites
- L3:1124 weakly suppresses
- L7:401 weakly suppresses
- L3:637 weakly suppresses
- L13:78 weakly excites
- L27:932 weakly suppresses
- L18:2401 weakly suppresses
- L8:283 weakly excites
- L4:319 weakly excites
- L13:961 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
- 鹄
- 为客户
- -sc
- mistakes
- ;break
- ate
- hacks
- -bo
- Ac
- -de
Suppresses
- ements
- ures
- [obj
- ilver
- Blond
- seulement
- qx
- ured
- CMD
- proj