unclear: acronym and term starts in lists
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. 24 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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
It fires on the first letter or word of a technical term or acronym, often right after 'the', a bullet or a list marker, such as 'NIM' or 'NURBS'. Pushing it down made the model talk about costs, which this neuron seems to hold back.
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.
-
• Describe the characteristics of capital expenditure⏎• Describe the capital budgeting process⏎• Understand and apply the following capital budgeting techniques:⏎o Payback period⏎o Net present value⏎o Internal rate of return<|endoftext|>Becoming a Reader:
-
hallmark traits of PANS includes an abrupt onset of OCD or severely restricted food intake, in conjunction with the acute onset of several neuropsychiatric symptoms described in the NIM
-
function for the state n = 3, ℓ = 0, and mℓ = 0 is where σ = r/a0 and a0 is the Bohr radius (5.29 × 10-11 m
-
specifications for 304 Plate and 304L Stainless Steel Plate are ASTM A-240,ASME SA -240 and A666.⏎results for this questionWhat is better 3
-
architectural projects and even special effects in movies. It is essentially the evolution of a merger between a graphical user interface and Non-Uniform Rational B-Splines (NURBS) geometry, and currently functions for all major platforms.⏎Become
-
and Christmas cactus (Schlumbergera bridgesii)⏎Take a pinch of fertilizer and put in the bottle or
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: what if we *plan* *grees* *grees* *grees* *grees* *grees* *grees* *grees* *grees* *grees* *grees* *g
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
entities to adjust their models accordingly and deliver optimized outputs and forecasts.⏎Digital twins are suitable for a wide range of applications, such as product lifecycle management (PLM), predictive maintenance of industrial machines, personnel training in virtual environments. In addition, digital
-
the total effect on Total Sales. However, if we want to know how the sales variance affects the profit, we use standard profit (absorption costing) or standard contribution marginal costing).⏎So, in order to calculate Sales Volume Variance, what
-
management framework means examining the entire range of business operations, identifying the strengths and weaknesses of current processes, and improving them. This includes individual workflows, administrative, support, and IT systems and the planning, production, and distribution of goods or services.
-
the standard price, standard profit or standard contribution.⏎This variance is quite important as it helps management assess how volume variance adversely affects the company’s performance. We use standard price if we want to assess how the variance affects total sales are.⏎We use
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
- L20:340 suppresses
- L18:3032 suppresses
- L20:1035 excites
- L7:2972 excites
- L11:1489 excites
- L6:2007 excites
- L4:2237 weakly excites
- L9:507 weakly suppresses
- L4:2629 weakly suppresses
- L8:1548 weakly excites
- L21:873 weakly suppresses
- L7:2960 weakly excites
- L9:2356 weakly excites
- L3:1006 weakly excites
- L1:2577 weakly suppresses
- L22:159 weakly suppresses
- L21:2721 weakly suppresses
- L22:2056 weakly suppresses
- L3:1558 weakly suppresses
- L6:1145 weakly excites
- L20:443 weakly excites
- L10:1937 weakly excites
- L9:364 weakly suppresses
- L10:1283 weakly excites
Feeds later layers
- L27:2264 suppresses
- L27:1263 weakly suppresses
- L25:488 weakly suppresses
- L26:738 weakly excites
- L25:992 weakly suppresses
- L27:1911 weakly suppresses
- L25:906 weakly excites
- L27:2952 weakly suppresses
- L25:2343 weakly excites
- L26:1543 weakly suppresses
- L27:1195 weakly excites
- L25:124 weakly suppresses
- L26:2065 weakly suppresses
- L27:1919 weakly suppresses
- L25:1347 weakly suppresses the word 'sales'
- L27:871 weakly excites
- L25:230 weakly excites
- L25:1184 weakly suppresses
- L26:1482 weakly suppresses
- L25:439 weakly excites
- L27:1595 weakly suppresses
- L27:836 weakly suppresses
- L27:2064 weakly suppresses
- L27:1959 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
- olds
- veyor
- ов
- chill
- vey
- �
- 冷水
- uler
- rá
- ظ
Suppresses
- 成本
- Cost
- cost
- 的成本
- cost
- costing
- Cost
- _cost
- Costs
- costs