zero digit inside a number
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. 10 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 '0' written in the middle of a number, such as in page numbers, p-values, dates, and counts. Turning it up made the model produce unrelated, wandering text.
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.
-
could pave the way for all-optical circuits (M Mützel et al 2002 Phys. Rev. Lett. 88 083601-1).⏎The light field of an interference pattern can
-
[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
-
of Muhlenberg College’s newest sustainability initiatives.⏎By: Grace Oddo '22 Wednesday, June 30, 2021 02:03 PM⏎Professor of Biology and Director of Sustainability Studies Rich N
-
9-0.965). Children were more vulnerable to rotavirus (p = 0.008, OR<|endoftext|>A Global Index of Sustainable Well-Being⏎Source: The
-
Central Business District as seen from the top of the Carlton Centre⏎|Municipality||City of Johannesburg|⏎|• Total||1.09 km2 (0.42 sq mi)|⏎|• Density||13,00
-
. Check out our full-length feature to find out more about these animals' astounding mental skills.<|endoftext|>- Cooking Time: 0⏎- Servings: 0⏎- Here are some recipes I have collected over the years and I am thrilled
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: which is the one thing you have. So, which is a thing. So, which is a thing. So, which is a thing. So, which is a thing. So, which is a thing. So, which is a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of Biblical Interpretation found in the catalog.⏎April 1, 2003⏎by Baker Academic⏎Written in English⏎|The Physical Object|⏎|Number of Pages||180|⏎Hermeneutics gives the reader
-
short-listed, especially since they’re volatile. With short-selling, volatility gives investors a chance to make big money.⏎Short-selling is an investment practice where you borrow stocks and sell them at a loss in hopes the price will drop. The purpose
-
WHO⏎The number of deaths from measles has fallen by 79 percent worldwide since 2000, thanks mainly to mass
-
the 1990s – when rising North Sea gas production and regulatory changes led to a massive increase in gas fired generating capacity – meant power emissions fell by 45 million tonnes or 22% over the decade. It’s
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
- L7:873 suppresses
- L9:977 suppresses the last digit of a decade year
- L9:657 excites
- L8:41 excites
- L8:20 excites
- L9:424 suppresses
- L3:651 excites zero in dates inside URLs
- L5:130 suppresses
- L8:318 suppresses
- L7:753 weakly excites
- L8:707 weakly suppresses
- L9:1531 weakly suppresses
- L4:326 weakly excites
- L9:1744 weakly suppresses
- L9:2319 weakly suppresses
- L0:208 weakly excites the digit zero inside codes or numbers
- L9:1032 weakly suppresses
- L4:1215 weakly excites
- L9:1348 weakly excites
- L8:660 weakly suppresses
- L7:2983 weakly excites
- L7:830 weakly excites
- L5:2102 weakly excites
- L9:135 weakly excites
Feeds later layers
- L18:726 suppresses
- L16:724 suppresses
- L11:553 suppresses
- L16:1333 suppresses 'two' after 'divided into' or 'there are'
- L12:713 suppresses
- L13:1479 suppresses
- L16:1420 suppresses
- L18:410 suppresses
- L16:1060 suppresses
- L14:1528 suppresses
- L19:723 weakly suppresses
- L11:2994 weakly excites
- L18:2831 weakly suppresses
- L22:332 weakly suppresses
- L18:1166 weakly suppresses
- L12:2434 weakly suppresses
- L16:2056 weakly suppresses
- L21:110 weakly suppresses
- L17:907 weakly suppresses
- L19:946 weakly excites
- L19:851 weakly suppresses comma inside numbers with thousands
- L11:502 weakly excites
- L20:1715 weakly suppresses
- L11:1427 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
- 興趣
- uja
- IEWS
- тра
- хот
- aly
- 投降
- 的兴趣
- obbies
- ��
Suppresses
- iteration
- looping
- rencontrer
- placeholder
- Iter
- -bit
- 舸
- cyan
- 环
- Diameter