second half of multi-word proper 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. 6 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 the piece of a name that finishes a well-known institution or person, like the 'etter' in Sloan-Kettering or 'Had' in Large Hadron Collider. When it was turned up, the model's writing did not visibly change in a clear way.
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.
-
.⏎"As the field broadens, the need for more counselors and more thorough training will also increase," says Lauren Scheuer, a genetic counselor at Memorial Sloan-Kettering Cancer Center in New York and a graduate of the Sarah Lawrence Program. "
-
hallmark of cancer and has a well-established role in aggressive B cell malignancies.⏎To find out more, Hans-Guido Wendel and colleagues from Memorial Sloan-Kettering Cancer Center in New York analyzed genomic data from two large groups of slow-growing
-
31-kilometre-long collider; like the Large Hadron Collider, it promises to recreate conditions moments after the big bang in an effort
-
requires policymakers to embrace a common goal and to stay the course for many years, shielding scientific workers from the shifting whims of politicians.⏎The National Aeronautics and Space Administration (NASA) is the largest individual sponsor of astrophysics and planetary
-
Daytona Beach, the WACS, the Women’s Auxiliary Corps, developed a training base. Noted African American educator Mary McLeod Bethune had lobbied President Roosevelt to set up a WAC training base
-
“I was far from convinced,” says Incandela, a professor at the University of California, Santa Barbara and the former spokesperson of the CMS experiment at the Large Hadron Collider.⏎For decades, scientists had searched for the elusive Higgs boson
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: "if not... then..." and so on... for the life of me... not knowing... what the heck... with all there is... no... where there is no... none of the rest... and so on... So, this is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, it would send a very fast message to the body to step back out of harm’s way. Similarly, if your stomach was empty and your body needed fuel,
-
the ip address. The number of port in computer networking is expressed in 16 bits, which means that there is a maximum of 65,536 ports on a computer. Assign the call coverage path for voice ports b7
-
way in which Windows configured the partition offset to attempt to achieve partition alignment.⏎Starting with these newer operating systems, the default partition offset is 1,048,576 bytes, exactly 1024 KB or 1
-
problem with hazardous smog, China seems dead serious on turning over a new leaf as we are now seeing a nation acknowledging the threat of climate change, 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
- L5:1899 strongly excites
- L5:2665 strongly suppresses
- L5:2136 strongly excites
- L5:2676 strongly excites
- L5:3063 strongly suppresses
- L5:2065 strongly suppresses
- L5:2708 strongly excites
- L5:2014 strongly suppresses
- L5:2971 strongly excites
- L5:3041 strongly suppresses
- L5:2206 strongly excites
- L5:2173 strongly excites "late" in dated phrases like "in the late 1980s"
- L5:297 strongly excites
- L5:2561 strongly excites
- L5:2452 strongly suppresses
- L5:2115 strongly excites
- L1:328 excites
- L3:1661 suppresses
- L5:2888 suppresses
- L0:1902 excites
- L2:1840 excites
- L2:693 suppresses
- L5:2618 excites
- L5:2295 suppresses
Feeds later layers
- L8:849 strongly excites
- L12:576 strongly suppresses 'of' in fixed phrases like table of contents
- L8:1017 strongly suppresses
- L9:716 strongly suppresses
- L14:1403 suppresses
- L10:129 excites
- L9:845 suppresses
- L15:2628 suppresses
- L10:678 suppresses
- L11:1384 suppresses
- L18:828 suppresses
- L16:882 suppresses
- L7:2370 suppresses
- L11:171 excites
- L15:1304 suppresses
- L20:452 excites
- L13:1513 suppresses
- L17:892 suppresses
- L10:952 excites
- L13:2328 excites
- L13:450 excites
- L19:2054 suppresses
- L19:1385 suppresses
- L18:819 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
- 因为他们
- ascript
- питания
- tics
- ibrate
- ervation
- cial
- خر
- 当前位置
- 农业大学
Suppresses
- ..."⏎⏎
- ..."⏎⏎
- lijk
- ...",
- 所
- ...',
- ENDED
- lu
- ...)⏎⏎
- "...