dollar sign before amounts
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 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 dollar sign that comes right before a price or amount of money, like "$0.05" or "$12,000". When it was turned up, the written text drifted into odd, unrelated rambling, so the steering results don't show a clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
), the nickel ($0.05), the dime ($0.10), the quarter ($
-
4-5 hours a day which are too warm for cow comfort. The estimated impact on milk production is 6 kg MS per cow over summer, or $12,000 for a 300-cow herd at
-
, tested the strength of common building materials such as pipes, and above-ground structures such as bridges and power lines. During the first test run at Cornell University’s $2.1 million earthquake lab, researchers simulated the effects of an earthquake on a
-
first to come up with these sorts of self-reliant devices. Back in 2015, for example, a separate team developed a tiny, $0.20 temperature sensor that didn’t require a battery or constant source of
-
defined, however some architectural considerations can be addressed and some conclusions can be reached.⏎Structure of Information⏎Consider these structures of data that can aid learning.⏎- The $M$ features of user linguistic response can be saved within a user session as a
-
, almost $8 for a gallon of gasoline and $6.50 for a gallon of
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: it's not a birth ritual in the sense that it's not in the birth pattern and not in the birth ritual in the ritual. It's not a birth in the birth ritual, it's not a birth in the birth pattern. It
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
slash ( `` ). Some common escape sequences are: - `n` Newline - Prints a newline character - `t ` Tab - Prints a tab character - `\ ` Backslash - Prints a backslash ( \ ) - `' ` Single quote
-
*) memchr (str, 'p', strlen(str));⏎printf ("'p' found at position %d.\n", pch-str+1);⏎printf ("'p' not found.\
-
Whether you’re a<|endoftext|>- Enter a word for the dictionary definition.⏎From The Collaborative International Dictionary of English v.0.48:⏎Nectarine \Nec`tar*ine"\ (n
-
$\quad+\quad$ or $\quad-$<|endoftext|>Design for a Mosaic⏎Albert Joseph Moore
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
- L19:329 strongly suppresses
- L22:326 excites
- L17:719 suppresses
- L21:1042 suppresses
- L21:588 excites
- L19:1040 suppresses
- L21:1374 excites
- L22:2028 excites
- L17:527 suppresses
- L18:724 suppresses
- L18:153 suppresses
- L21:2508 suppresses
- L20:934 weakly excites
- L8:16 weakly suppresses
- L20:594 weakly suppresses
- L19:58 weakly excites
- L2:28 weakly excites
- L3:204 weakly excites
- L18:1351 weakly suppresses
- L20:496 weakly excites
- L17:2329 weakly suppresses
- L22:2107 weakly excites
- L1:2340 weakly suppresses
- L18:569 weakly excites
Feeds later layers
- L26:473 strongly excites
- L26:245 excites
- L24:54 excites
- L25:391 excites
- L27:171 excites
- L27:430 excites
- L25:630 excites
- L25:560 excites
- L27:1024 excites
- L24:458 excites
- L27:1480 excites
- L24:218 excites
- L24:256 excites
- L25:524 excites
- L25:487 suppresses
- L25:276 excites
- L25:380 excites
- L25:1952 suppresses
- L24:117 weakly excites
- L24:1601 weakly excites
- L25:113 weakly excites
- L26:540 weakly suppresses
- L27:1241 weakly excites
- L24:653 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
- \
- \%
- \v
- \t
- \d
- ^\
- \r
- \
- ^{
- (\
Suppresses
- อำนาจ
- ($("#
- =$("#
- �
- **)
- 在现场
- ="--
- impres
- ($(".
- (MouseEvent