equals sign or 'is' in formulas
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. 14 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 equals sign or the word 'is' in maths and code, right after a variable name, as in 'vx = ' or 'Step angle ='. Turning it up pushed the writing toward odd number-heavy or unrelated text.
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.
-
= vsin(theta) Immediately after take off : vx = 2cos25 = 1.8126 vy =
-
2, where k is the number of digits of 2n. The number of digits can be estimated as n log102. Thus, the probability is about cxn, where c = (10/9)2 ≈
-
angles of 30, 15, 5, 2.5, 2, and 1.8° per step.⏎Step angle = 360°/No of steps per revolution⏎Where m = Number
-
DataTableCollection.⏎Private Sub GetTableByName() ' Presuming a DataGrid is displaying more than one table, ' get its DataSet. Dim thisDataSet As DataSet = CType(DataGrid1.DataSource, DataSet) ' Get the DataTableCollection. Dim tables
-
I’ll keep it that way. That just means that I have to make sure that I use gravity as a negative number, since gravity always acts down.⏎vf = vi + at⏎= 56.3m/s + (-9.
-
function⏎unorm = sqrt(u(1)*u(1)+u(2)*u(2)); %calculate intermediate value⏎udot(1) = -u(1)/R + u(2) - un<|endoftext|>Thanks for
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 idea of a modern-day "evidence for the universe that we can say it's right at the core of our universe when the universe is at the end of the universe when the universe is at the time of our existence" was a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Grid is displaying more than one table, ' get its DataSet. Dim thisDataSet As DataSet = CType(DataGrid1.DataSource, DataSet) ' Get the DataTableCollection. Dim tablesCollection As DataTableCollection = thisDataSet.Tables ' Get a specific table by
-
⏎t0=0; %set the initial time to zero⏎tend=1; %set the final time to 100⏎
-
size) to determine the next point. For example, if the gradient magnitude is 2.5 and the learning rate is 0.01, then the gradient descent algorithm will pick the next point 0.025 away from
-
+ 314⏎Are the denominators equivalent? Yes, they are.⏎Now we can add the cardinal numbers and the numerators. First the cardinal numbers. That is 2 + 3 = 5. And then
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
- L13:1147 suppresses
- L13:1988 excites
- L12:1865 suppresses
- L12:1973 weakly excites
- L12:2101 weakly suppresses
- L13:1591 weakly suppresses
- L12:1254 weakly suppresses
- L10:1924 weakly suppresses
- L13:2345 weakly excites
- L12:2781 weakly suppresses
- L6:12 weakly excites
- L13:1741 weakly excites
- L13:2954 weakly excites
- L12:737 weakly suppresses
- L13:1561 weakly excites
- L10:1625 weakly suppresses
- L13:1137 weakly suppresses
- L2:808 weakly excites
- L13:2424 weakly excites
- L5:1566 weakly suppresses
- L13:78 weakly excites
- L8:74 weakly suppresses
- L13:960 weakly excites
- L12:1175 weakly excites
Feeds later layers
- L15:2583 suppresses
- L15:1854 suppresses
- L16:1514 suppresses
- L16:2786 excites
- L16:789 excites
- L15:1642 excites
- L16:1995 excites
- L16:1207 excites
- L16:1219 excites
- L15:91 excites
- L15:1977 suppresses
- L17:3024 suppresses
- L15:2432 weakly suppresses
- L15:1795 weakly excites
- L18:2631 weakly suppresses
- L19:2068 weakly excites
- L15:2554 weakly suppresses
- L15:2211 weakly excites
- L15:1702 weakly excites
- L18:1537 weakly suppresses
- L18:115 weakly excites
- L16:616 weakly suppresses
- L16:702 weakly excites
- L16:2971 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
- ESH
- amiento
- uble
- uario
- стрел
- изации
- _follow
- entifier
- ahan
- phép
Suppresses
- total
- 总
- totaled
- .total
- 总投资
- curring
- 保持
- ,total
- total
- 总体