"the" right after "without"
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. 2 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 lights up on the word "the" when it comes straight after "without", as in "without the protection" or "without the owner's knowledge". When it was turned up, the model's text did not visibly change in a way that connects to this idea.
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.
-
, and spectacular towers offering strategic vantage points were just some of the mechanisms used back in the day.⏎Now try imagining your business as a castle, just without the adequate protection. It’s undeniable that external threats are targeting businesses of all sizes,
-
impact."⏎As a result, he says, "Most houses in North Carolina that are oceanfront will not make it through a 30-year mortgage" without the government spending money to improve and sustain the area's infrastructure. "The sea level
-
common purpose. Conversely, traditional work teams are highly structured and are sometimes confused with the work group. Teamwork can be difficult to establish within a traditional team without the identification of a common purpose to which all workers can agree and commit.⏎The self
-
dead at my feet than guilty of a mortal sin.” His biographers have written of the long hours he spent in prayer, fasting, and penance, without the knowledge of his people.⏎The French king was an avid lover of justice, who
-
appreciate the other side of e-cigarettes. For those unwilling or unable to give up nicotine, e-cigarettes offer something akin to the experience of smoking a real cigarette without the same concentration
-
a minimum, they should be cognizant that it is malicious software that collects personal information, usually with the intent of sending that information to third parties, without the owner's knowledge or consent. If anyone would like to learn the historical, technical
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 have been thinking about lately, and I have been thinking about... But this is not just a new month, but this is not just a new year, but this is not just a new week. I have been thinking about lately,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Link for teaching and learning in the New Year. The ideas are for those who are just starting out, or for those who
-
-nine cases cured and a large number "permanently arrested and greatly improved." Von Ruck continued<|endoftext|>Tamales are traditional Mexican foods often served at New Year’s Eve or Christmas Eve dinners. They are steamed bundles of corn dough stuffed
-
vegetables and less processed foods combined with Americans’ favorite past-time of adopting restrictive, hard-to-maintain, weight-loss inducing diets at the turn of the New Year, creates a perfect storm for a surge in interest and popularity of the raw food
-
of the Jewish high holidays and the civil year on the Jewish calendar. It’s a good time to wish your Jewish friends leshana tova (happy New Year), to send greeting cards, and perhaps plant trees in Israel in their honor.⏎
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
- L1:1796 weakly excites
- L0:178 weakly suppresses
- L0:1301 weakly suppresses
- L1:59 weakly excites 'the' after a verb or preposition
- L0:1224 weakly suppresses
- L0:253 weakly suppresses
- L1:24 weakly excites
- L1:2689 weakly suppresses
- L1:305 weakly suppresses
- L1:2188 weakly suppresses
- L1:2611 weakly suppresses
- L1:2074 weakly suppresses
- L1:2831 weakly excites
- L1:2119 weakly excites
- L0:774 weakly suppresses
- L0:510 weakly suppresses
- L1:2518 weakly excites
- L0:2340 weakly suppresses
- L0:2686 weakly suppresses
- L0:2999 weakly excites
- L1:2919 weakly suppresses
- L0:2706 weakly excites
- L1:2843 weakly excites
- L1:2041 weakly suppresses
Feeds later layers
- L16:2621 weakly excites
- L18:2336 weakly suppresses
- L23:1034 weakly suppresses "set" before out, off or up
- L8:2461 weakly excites
- L20:2639 weakly suppresses
- L17:446 weakly excites
- L18:95 weakly suppresses end of a document marker
- L15:2600 weakly excites
- L17:1909 weakly excites
- L9:1532 weakly suppresses
- L5:2537 weakly excites
- L11:1012 weakly excites
- L13:511 weakly suppresses
- L11:289 weakly excites
- L11:421 weakly suppresses
- L6:1209 weakly excites
- L14:2573 weakly excites
- L4:2171 weakly suppresses
- L15:2493 weakly excites
- L3:1663 weakly suppresses
- L7:2817 weakly excites
- L15:2181 weakly suppresses
- L11:1471 weakly excites
- L8:478 weakly 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
- @foreach
- ��
- 写字
- 资本市场
- OWNER
- astr
- acres
- 率达
- urai
- ':'
Suppresses
- énom
- wares
- rome
- ripsi
- genu
- visor
- körper
- cludes
- ideshow
- roids