'a' before an object noun
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. 20 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
Fires on the word 'a' when it comes right after a verb or preposition and a noun is about to follow, such as 'when requesting a correction'. Turning it down made the model repeat 'a a a', so the output is mostly broken.
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.
-
VECE20|⏎|Order Information:||Web: http://www.tandfonline.com/pricing/journal/VECE20|⏎When requesting a correction, please mention this item's handle: RePEc:taf:j
-
objective.⏎Student data on learning objectives also generate a learning estimate that appears in the instructor's Learning Dashboard. The learning estimate predicts how well a student will perform on a learning objective.⏎Summative assessments are assignments that assess mastery of material. Scores
-
adjustments to the apparatus are being made, the hood should be kept closed, with vertical sashes down and horizontal sashes closed, to help prevent the spread of a fire, spill, or other hazard into the laboratory. Sliding sashes should
-
based on the borrower's income, expenses, and credit history. The results of a stress test can impact the amount a borrower is able to borrow for a mortgage and the interest rate they receive.⏎Debt Servicing Ratio is a measure
-
in this study are freely available to the broader research community, and researchers can use these data to study the cells in greater depth and determine how phosphorylation events determine a cell's fate," said Dr. Laurence M. Brill, senior scientist at
-
Angle in End Milling⏎End mills are used to form shapes and holes during milling and can cut a variety of materials in several directions. The end mill is generally described as a positive
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: a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
prove its identity as a unique physiographic region and as a region that is beneficial to grape growing. Some of the criteria used by the federal government to determine whether an area meets this standard include distinct climate, soil type, and elevation and other
-
incident to two nodes. We define adjacent nodes of a given node as alternate nodes of incident edges. Adjacent nodes are sometimes referred to as neighbors. The degree of a node is the number of its incident edges or
-
the exact value of the number. Once you have the answer, you can move onto the next step in interpreting the result.⏎The first step in finding the percentage of a
-
percent. It’s easy to do this with the calculator. You can use a different method to find the percentage of a number.⏎Another way to calculate the percentage of a number is to divide the number by its maximum possible value. You can do
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
- L19:2972 suppresses
- L19:684 suppresses
- L18:692 excites
- L19:2269 suppresses
- L19:654 excites
- L18:1538 suppresses
- L18:3050 suppresses
- L17:1527 excites
- L18:2323 suppresses
- L18:2947 excites
- L18:2404 suppresses
- L17:817 suppresses
- L16:2332 suppresses
- L19:1352 suppresses
- L14:2629 suppresses
- L19:2274 excites
- L18:376 weakly suppresses
- L16:2363 weakly excites
- L19:626 weakly excites
- L19:1710 weakly excites
- L18:406 weakly excites
- L17:1429 weakly excites
- L19:1305 weakly suppresses unclear: verbs after a noun or 'that'
- L0:1826 weakly excites
Feeds later layers
- L22:103 strongly excites
- L25:45 strongly excites
- L27:797 strongly suppresses
- L21:1457 strongly excites
- L25:1586 strongly suppresses
- L27:663 strongly suppresses
- L24:823 strongly suppresses
- L26:484 strongly suppresses
- L24:2538 strongly suppresses
- L27:709 suppresses
- L25:1132 excites
- L26:213 excites
- L27:231 excites
- L24:488 excites
- L25:871 suppresses
- L27:1044 suppresses
- L21:1720 excites
- L27:173 excites
- L21:775 excites the article 'a' before a noun
- L27:2396 suppresses
- L23:1220 weakly suppresses
- L23:119 weakly suppresses
- L21:592 weakly excites
- L25:568 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
- esse
- inished
- icated
- 各
- imer
- each
- -Saharan
- rete
- IVED
- ihan
Suppresses
- a
- an
- 一个
- 一种
- →a
- something
- 一项
- 一款
- 一種
- 的一种