'of' inside organization 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. 16 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 'of' when it appears inside the name of an institution, like 'Department of Natural Resources' or 'Georgia Institute of Technology'. It switches off on 'of' in phrases like 'Moore's law' or 'Newton's law of'. Turning it up or down did not give a clear theme in the 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.
-
benefit to the host via reduction of antigen load from the gut.”⏎Intermittent Fasting May Provide Comparable Health Benefits to Calorie Restriction<|endoftext|>The Minnesota Department of Natural Resources announced Tuesday that an eaglet has hatched in the nest of two
-
and 4. Shell damage may be due to predation by fish (i.e. common carp or freshwater drum). Top photo by Mike Davis, Minnesota Department of Natural Resources; bottom photo by Gary Wege, U.S. Fish and Wildlife
-
learning that are essential for all students. They are the ideas, concepts, and skills that provide a foundation on which subsequent learning may be built. The Arizona Department of Education’s Virtual Learning Hub is a resource for teachers and families to assist them as
-
processing:⏎3. Masks that have to be subjected to a test for temperature resistance at 70°C over 24 hours before reprocessing:⏎Federal Ministry of Health. Use of protective masks in healthcare facilities. Paper from the crisis team with
-
of the Georgia Institute of Technology said in a video about the work.⏎Kemp and his team programmed a
-
paper is featured this month by the American Psychological Association on the Office of Ethnic Minority Affairs' African American Heritage Month website, found at: http://www.apa.org/pi/oema/resources/ethnicity-health/african-american/index.aspx⏎"So
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: it's the one with the three different states. It's the one with the three states of the one way? Or how exactly? So, in case, I can be divided into three states? Or how? Also, how does that
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and is known only to the Butonian people on the Pulau Makassar. The study used the symbolic interaction theory by Herbert Blumer. The identification of six informers was deemed viable. A
-
the mathematical tools needed to compute the motions of these bodies. Given initial positions and velocities, we can calculate the forces acting on each object, using Newton’s law of universal gravitation. From the forces we can determine accelerations, and then update
-
chip performance—measured by the number of transistors on a chip—doubles every 18 months.⏎It is no secret that Moore’s law, which has driven progress in the computer industry and the economy for decades, no
-
⏎Your credit report is your window into your ID security. The Fair and Accurate Credit Transactions Act, passed by the Federal government in 2003, mandates that each of the major credit bureaus supply consumers with a free copy of
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
- L8:2050 suppresses
- L3:616 suppresses
- L3:2793 excites
- L12:3035 excites
- L10:2129 weakly excites
- L12:708 weakly suppresses
- L15:1786 weakly suppresses
- L6:2394 weakly suppresses
- L10:448 weakly suppresses
- L3:1835 weakly excites
- L5:1132 weakly excites
- L4:1418 weakly suppresses
- L11:2177 weakly suppresses
- L4:1200 weakly suppresses
- L4:2616 weakly suppresses
- L7:1885 weakly suppresses
- L14:2147 weakly suppresses
- L11:1268 weakly excites
- L7:1063 weakly excites
- L15:1427 weakly suppresses
- L10:1070 weakly suppresses
- L15:718 weakly suppresses
- L11:1874 weakly suppresses
- L10:1755 weakly suppresses
Feeds later layers
- L19:1630 excites
- L20:706 excites
- L23:1696 weakly suppresses
- L19:2840 weakly excites
- L18:2599 weakly excites
- L20:2012 weakly excites
- L24:1241 weakly excites
- L17:779 weakly excites
- L17:110 weakly excites
- L26:1629 weakly excites
- L21:2450 weakly excites
- L18:2941 weakly excites
- L22:2320 weakly suppresses
- L17:3014 weakly excites
- L18:2868 weakly excites
- L17:1136 weakly excites
- L17:2491 weakly suppresses
- L21:1344 weakly excites
- L23:2585 weakly excites
- L20:2809 weakly suppresses
- L22:1813 weakly excites
- L20:606 weakly suppresses
- L20:1247 weakly excites
- L17:352 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
- 用户提供
- 尊敬
- 用户的
- 箭
- mounted
- jumper
- 兴趣
- finish
- 击杀
- 👤
Suppresses
- plies
- og
- ocate
- icles
- ulate
- ulates
- ogi
- ibbean
- uento
- Framework