digits inside numbers
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. 12 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 a digit that sits inside a number, such as the 0 in '£0.00', the 0 in a storm code like '09L', or the 1 in '1000'. It is strongest on single digits in counts and codes. When turned up, the model's output did not show a clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
2013 - Tropical Storm Humberto (09L) off West Africa⏎On September 11, 2013 Tropical Depression 09L had strengthened enough to become the first storm of the year to reach hurricane
-
[The Daily Journal]<|endoftext|>Should You Ask Your Children to Apologize?By Craig Smith | September 23, 2016 | 0 comments⏎When kids say sorry, are they learning a lesson or just parro
-
often referred to as the 16-bit era, and the previous generation as the 8-bit.⏎About Gaming Consoles⏎1 reply to this topic⏎0 user(s) are
-
digit of one of the numbers, and then sum the results together.⏎There is a single positive integer T on the first line of input (equal to about 1000). It stands for the number of expressions to follow. Each expression
-
to see the full course over view and dates of courses running in your area.⏎You have 0 courses in your shopping cart⏎Current total : £0.00 + VAT⏎Join our mailing list to receive our tech
-
reading this topic⏎0 members, 0 guests, 0 anonymous users<|endoftext|>The JCTC applauds the action by
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: I can use a certain level of information in a certain way, but I can make sure that the user will have to follow certain conditions. In other words, I can make sure that the user will have to be informed of the rules,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Children and Adolescents (2011). Expert panel on integrated guidelines for cardiovascular health and risk reduction in children and adolescents: Summary report. Pediatrics, 128(S
-
be well represented in the efforts currently under way to elucidate the molecular basis of the genetic susceptibility to alcoholism.(JAMA. 1992;268:1877-1882)<|endoftext|>The Imag
-
ar abscess, mastoiditis, and rheumatic fever in children: time trend analysis. BMJ. August 6, 2005;331:328–9.⏎Used with permission from
-
12). Social connections and the persuasiveness of viral campaigns in social network sites: Persuasive intent as the underlying mechanism. Journal of Marketing, 18(1), 39-53. http://dx.doi
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
- L6:2938 suppresses
- L10:800 excites
- L11:1601 excites
- L11:1546 excites
- L9:2890 excites
- L10:810 suppresses
- L11:313 weakly excites
- L11:300 weakly suppresses
- L0:1301 weakly suppresses
- L2:1703 weakly suppresses
- L4:3023 weakly suppresses
- L11:1153 weakly suppresses
- L1:2536 weakly excites
- L9:726 weakly suppresses
- L10:1705 weakly excites
- L9:2211 weakly suppresses
- L3:464 weakly excites
- L10:509 weakly excites
- L11:18 weakly suppresses
- L11:553 weakly suppresses
- L10:926 weakly suppresses
- L11:1716 weakly excites
- L2:1698 weakly excites
- L8:2337 weakly suppresses
Feeds later layers
- L14:542 strongly suppresses
- L17:259 suppresses
- L19:395 suppresses
- L16:256 suppresses number right after 'age of' or 'ages'
- L15:646 suppresses
- L16:2735 suppresses
- L16:3057 suppresses
- L17:533 suppresses
- L14:759 suppresses
- L16:582 suppresses
- L18:170 suppresses
- L13:999 suppresses
- L25:182 suppresses
- L14:311 suppresses
- L15:1900 excites
- L16:286 suppresses
- L15:364 suppresses
- L19:672 suppresses
- L16:253 suppresses
- L16:210 excites
- L24:197 suppresses
- L21:175 suppresses
- L19:851 suppresses comma inside numbers with thousands
- L20:1567 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
- powered
- lifted
- Code
- extremism
- intage
- lifting
- ismatic
- ثور
- shaking
- typeof
Suppresses
- ços
- olini
- apixel
- erten
- ert
- ertain
- olin
- 年由
- 日正式
- erset