blank token before a number or after a count
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 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
The neuron reacts to an empty spot in the text, often in web page footers like "3 Views [blank] 20 Downloads" or between two numbers. It is hard to say more than that. Turning it up or down made the model's writing shift topic without a clear pattern.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
3: 25. doi:10.1023/A:1005462530658⏎- 896 Downloads⏎Using recently available data from China's Disease Surveillance Points system
-
well as improve their athletic skills. They work cooperatively playing soccer, badmitton, frisbee, bean bag bowling and other games.⏎17 Views 20 Downloads<|endoftext|>In Key Stage 1 the children will:⏎The five enquiry
-
explore the Orange Revolution. In this Orange Revolution lesson, students examine the climate of the Ukrainian protest and investigate the role of Canadian government in the protest.⏎3 Views 2 Downloads<|endoftext|>The environment contains those factors ‘surrounding the organization’.
-
paper, with margins; letterpress text on verso; platemark: 422 x 531 mm; total: 518 x 638 mm; watermark: Atlas holding the celestial spheres; slightly frayed
-
extraordinary 33 years in average life span over the past century. A baby born today can expect to live an average of 78 years, compared with 45 years back in 1900. But those gains have not
-
verb tense review worksheet, students are given sentences and must find the mistake. Students correct the mistakes when they find them using their knowledge of verb tense.⏎3 Views 28 Downloads<|endoftext|>Blank Electoral College Map 201
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 first thing to understand is that the first thing to understand is that the first thing to understand is to understand that the first thing to understand is to understand that the first thing to understand is to understand that the first thing to understand is to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
"documentation style"?⏎A documentation style is a standard, agreed-on method for creating citations. It provides formats for citing differnet information resources within the body of the paper, for listing different information types on the Works Cited page, and even
-
fascinating historical perspectives, as well as covering the role of the Irish language in Northern Ireland's more recent past. But the main body of the book is based on Malcolm's extensive and detailed research into the attitudes of young Protest
-
345 Can we simplify? No, that's not possible.⏎The answer is 315 + 35 = 345⏎Example 2: 214
-
of eqns.⏎%for any input values t, y⏎ydot = -10*y %compute rhs⏎The function needs to be tested. To do this, issue the command⏎q = rhs(0.2,
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
- L5:53 weakly excites
- L3:1865 weakly suppresses
- L5:2407 weakly excites
- L2:70 weakly excites
- L4:53 weakly excites blank slot before a number
- L4:72 weakly suppresses
- L0:919 weakly excites
- L1:1674 weakly excites
- L0:21 weakly suppresses last digit of years like 1970
- L5:1195 weakly excites
- L5:46 weakly excites
- L0:1938 weakly suppresses
- L5:50 weakly excites
- L1:260 weakly excites
- L5:1939 weakly suppresses
- L3:2140 weakly suppresses
- L5:1264 weakly excites
- L3:230 weakly suppresses
- L2:155 weakly suppresses
- L5:322 weakly suppresses
- L3:1661 weakly suppresses
- L5:1761 weakly excites
- L5:285 weakly excites
- L5:932 weakly excites
Feeds later layers
- L7:2 strongly suppresses
- L10:17 strongly suppresses
- L7:24 strongly excites
- L11:43 strongly suppresses
- L9:8 strongly excites
- L11:41 strongly excites
- L11:518 excites
- L13:41 excites the word 'the' after a lead-in phrase
- L14:1962 suppresses
- L7:13 suppresses 'the' after 'member of' or 'one of'
- L12:130 excites
- L9:87 excites
- L14:1267 excites
- L10:42 suppresses
- L14:2272 excites
- L8:98 suppresses
- L10:146 weakly suppresses
- L7:184 weakly suppresses
- L18:505 weakly suppresses
- L16:1335 weakly suppresses
- L12:267 weakly suppresses
- L15:764 weakly suppresses
- L21:359 weakly excites
- L15:1815 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
- noon
- ateurs
- жа
- usan
- urar
- .AddInParameter
- haul
- ateur
- 长征
- Ữ
Suppresses
- ip
- ertz
- ip
- butt
- cip
- 种
- bonded
- posit
- orneys
- vip