the 9 in years like 19xx
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. 22 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 digit 9 in the middle of a year that starts with 1, such as 1983 or 1948. Its negative side is less clear, mostly showing up on words like 'paying' and 'no' in ordinary sentences. Turning it up did not produce any obvious theme in the output.
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.
-
. The process put in at home and inexpensive. It will require a bit of specialized knowledge and a few tools.⏎The first computer virus was developed in 1983 by Fred Cohen, a mentor at the School of New Dreamland.
-
prove it. Many conspiracy theories regarding secret chambers and passageways within the pyramids exist to this day.⏎These conspiracy theories became even more widespread in 1993, when Dr. Z<|endoftext|>Our cataloguer and book reviewer Jo takes
-
within Islam, but the majority in Iran and Iraq. Belief in the messianic Twelfth Imam was a key component of the Iranian Revolution of 1979
-
, or whether it’s nitrogen and phosphorus – will make it to the sea. And that means that probably water quality will<|endoftext|>Harlem riot of 1935⏎The Harlem riot of 1935 was Harlem's first
-
0 antibody can cure multiple types of cancer and prevent spontaneous genetically driven cancers.<|endoftext|>Do You Know?⏎National Guard Bureau⏎WASHINGTON, March 5, 1998 Our National Parks⏎
-
that the UK must reduce the amount of gold in its sovereigns, albeit by<|endoftext|>Oil Crisis and consolidation (1974–1988)⏎The first mild recession in the Federal Republic
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 first part of the course of a class of students. A class of students who will eventually leave a class with the last member, the last class out of the class of students, a class that will have a member who will leave a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
score?⏎Credit scores help lenders decide whether to lend money, how much to lend and how much interest to charge. By not knowing their credit score, people risk paying more interest when they take out a loan and may face limits on the amount they
-
expression. Digital space offers an affordable, accessible, and, if desired, anonymous, soapbox far freer than its terrestrial counterparts. So moving there is a no-brainer, as natural as water running downhill.⏎But the question that political scientists
-
)⏎- Mixed facts practice pages⏎- Answer keys⏎GRAB THE BUNDLE! It will get students EXCITED TO PRACTICE SUBTRACTION! No more boring flashcards over and over. Simply print and students are ready to engage
-
eat more healthy meals 75-80% of the time in a week, that’s 16 meals that are healthy. This way there is no feeling of failure if you aren’t eating an entirely healthy meal every time you eat
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
- L21:1435 suppresses
- L19:1301 excites
- L8:2024 weakly excites
- L18:2776 weakly excites
- L20:919 weakly excites
- L19:2792 weakly suppresses
- L4:42 weakly excites
- L7:71 weakly suppresses
- L18:311 weakly suppresses
- L15:2101 weakly excites
- L7:2895 weakly excites
- L5:1177 weakly excites
- L8:2458 weakly excites
- L10:23 weakly suppresses
- L9:345 weakly excites
- L15:2189 weakly excites
- L14:2519 weakly excites
- L12:2401 weakly suppresses
- L18:2561 weakly suppresses
- L17:292 weakly excites
- L8:1990 weakly excites
- L13:63 weakly excites
- L17:978 weakly suppresses
- L15:2754 weakly excites
Feeds later layers
- L23:3018 weakly excites
- L23:3042 weakly excites
- L23:1666 weakly excites
- L25:2781 weakly excites
- L27:1694 weakly excites
- L25:1373 weakly excites
- L23:1068 weakly excites
- L24:2817 weakly excites
- L27:1736 weakly suppresses
- L26:1273 weakly excites
- L26:2025 weakly suppresses
- L23:1622 weakly excites
- L26:2351 weakly excites
- L27:1480 weakly suppresses
- L24:514 weakly excites
- L23:978 weakly excites
- L24:2017 weakly suppresses
- L25:2909 weakly suppresses
- L23:787 weakly suppresses
- L24:2609 weakly suppresses
- L24:2740 weakly excites
- L25:445 weakly suppresses
- L23:587 weakly excites
- L24:2630 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
- рест
- Wear
- Characters
- Characters
- elan
- aged
- 唧
- итет
- 惺
- itat
Suppresses
- {name
- cole
- {k
- ucket
- cole
- {[
- {n
- Cole
- MARK
- cur