the word 'brake' and braking
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. 23 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
This neuron lights up on the word 'brake' (and 'braking'), especially in car and machine text like brake pads, rotors and pedals. Turning it up made the model's text drift to unrelated topics like chemistry or math, so the steering results don't show the brake theme.
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.
-
Corvette Stingray (C2).⏎These brakes offer better stopping performance than comparable drum brakes, including resistance to "brake fade" caused by the overheating of brake components, and are
-
the brake pads as they wear, a technician must make certain that the brake rotors are within certain standards to stop the vehicle operating in a smooth and reliable
-
brakes are less effective). Unlike a drum brake, the disc brake has no self-servo effect and the braking force is always proportional to the pressure placed on the braking pedal or lever.⏎Many early implementations for automobiles located the brakes on the inboard
-
posites (including carbon, kevlar and silica), is connected to the wheel and/or the axle. To stop the wheel, friction material in the form of brake pads (mounted on a device called a brake caliper) is forced mechanically,
-
asbestos in heat resistant products like heat–resistant pads, asbestos gloves, and woven into cloth coverings and protective blankets.⏎There was extensive use of asbestos in brake pads and shoes, as well as asbestos clutch discs and linings. For mechanics
-
give the car locomotion are turned in reverse during braking turning them into generators, returning the power to the batteries to regenerate them.⏎In this system the amount of brake pedal pressure denotes how much braking power is manual versus how much electrical energy is rec
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 value of a sample (n) in the study. I have been concerned about the value of the sample (n) and the value of the sample (n) in the study. I have been concerned that the value of the sample
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
mRNA, miRNA, and other small RNAs were extracted from specimens using TRIzol reagent. Qiagen's miScript System was used to synthesize cDNA and perform qPCR (
-
were collected for subsequent analysis8.⏎Quantification of miRNA levels in prostate tumor samples. The main steps in qPCR analysis of tumors are: Total RNA extraction, cDNA synthesis, and detection of qPCR products using miRNA-specific primers
-
the next generation of biomarkers, therapeutic agents and/or targets.⏎Methods used to investigate miRNA expression include SYBR green I dye- based as well as Taqman-probe based qPCR. If miRNAs are to be effectively used
-
technology used to take accurate rake angle measurements.⏎- Rake Angle⏎- Positive and Negative Rake Angle in End Milling⏎- Relationship Between Cutting Tool Angle, Rake Angle, and Relief Angle⏎- Problems in Conventional Rake Angle
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
- L22:167 excites
- L10:566 excites
- L20:410 excites
- L20:1023 excites
- L19:1158 weakly suppresses
- L1:719 weakly excites
- L15:1854 weakly excites
- L10:1092 weakly suppresses
- L5:1735 weakly suppresses
- L21:851 weakly excites
- L22:3029 weakly excites
- L9:720 weakly suppresses
- L10:2990 weakly suppresses
- L21:865 weakly suppresses
- L21:2094 weakly excites
- L19:1324 weakly suppresses
- L21:2184 weakly excites
- L5:208 weakly suppresses
- L10:1618 weakly suppresses
- L18:1460 weakly suppresses
- L1:951 weakly excites
- L9:3026 weakly excites
- L6:2295 weakly suppresses
- L14:2968 weakly excites
Feeds later layers
- L24:2907 weakly suppresses
- L27:2726 weakly excites
- L24:1445 weakly suppresses
- L25:1604 weakly excites
- L24:2751 weakly suppresses
- L27:888 weakly excites
- L26:2763 weakly excites
- L25:923 weakly excites
- L26:667 weakly excites
- L24:2255 weakly suppresses
- L27:41 weakly excites
- L24:2723 weakly excites
- L26:1248 weakly excites
- L24:499 weakly suppresses
- L26:412 weakly excites
- L24:2659 weakly excites
- L25:1819 weakly excites
- L26:762 weakly excites
- L24:2205 weakly excites
- L25:3068 weakly excites
- L27:2217 weakly excites
- L24:3027 weakly excites
- L26:2989 weakly suppresses
- L25:2559 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
- dryer
- سطين
- inidad
- bson
- lining
- ева
- 扶
- .EventArgs
- nes
- .Writer
Suppresses
- imenti
- digest
- olson
- 国王
- безопас
- guards
- 札
- 死去
- ipur
- issen