"first" in "for the first time"
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. 24 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 word "first" in the phrase "for the first time". When turned up, the model started repeating the word "time".
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.
-
was set up in the country just over a decade ago that has revolutionised African banking, and given access to financial services to millions of the poorest people for the first time. M-Pesa<|endoftext|>It has been said that that major historic events are
-
a, and the festive season in full swing, the spectre of an untreatable sexually transmitted infection looms over us - and our teenagers - for the first time in a generation.⏎Since penicillin was first used to treat gonorrh
-
made a mistake has never tried anything new.” ~ Albert Einstein⏎When we’re small children, our mistakes are applauded. Our falling-down attempts to walk for the first time are cheered by our parents, giving us the courage to get up and try
-
a new mass murder, the Holocaust, was unfolding before his eyes. Seventy years ago this autumn, Lemkin used the word “genocide” for the first time, in his book ‘Axis Rule in Occupied Europe.’ The 7
-
disciplines on the road to “the wolf’s lair.”⏎The result is an unsparing history of the German resistance to Hitler—one where the players emerge for the first time as real people with complex motives and
-
970's when President Nixon visited mainland China for the first time. Today, that may not seem like such a big deal. But back
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 effect of a medical diagnosis on the way we look at the health of our bodies and the health of our daily lives. After a medical diagnosis, someone is faced with the reality that their health is no longer the same as before. The
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the tar sands.⏎The Alberta tar sands has dominated headlines and television news since a Greenpeace Canada delegation arrived in Oslo Monday. Two of the delegates, Andrew Nikiforuk and Dr. John O'Connor, have raised concerns about the environmental,
-
by attacking the large corporation E Corp.⏎Elliot Alderson, its a young man living in New York City, who works at the
-
becoming an increasingly popular alternative to lawsuits. As a result,<|endoftext|>10 / The Natural World⏎|Artist / Origin||⏎Albrecht Altdorfer (German, 1480–1538)⏎
-
François Mitterand who, in 1954, defiantly pronounced that ‘L’Algérie, c’est la France’. Nicolas Sarkozy, on the other hand,
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
- L22:2238 suppresses
- L22:1195 suppresses
- L22:1956 suppresses
- L20:1894 excites
- L18:2696 suppresses
- L20:2636 excites
- L20:3014 excites
- L19:2098 excites
- L19:2739 suppresses
- L22:2916 suppresses
- L20:750 suppresses unclear: chinese academic text, 'when' loops
- L19:2766 excites
- L5:629 weakly suppresses
- L17:2750 weakly excites
- L22:2445 weakly suppresses
- L14:1102 weakly suppresses
- L21:1273 weakly excites
- L19:2073 weakly suppresses
- L17:1416 weakly suppresses
- L20:1199 weakly excites
- L17:93 weakly excites
- L16:94 weakly suppresses
- L18:2260 weakly excites
- L23:1549 weakly excites
Feeds later layers
- L27:836 strongly excites
- L25:565 strongly excites
- L26:2065 strongly excites
- L26:2567 strongly excites
- L27:1639 excites
- L26:549 weakly excites
- L25:281 weakly excites
- L26:528 weakly excites
- L25:770 weakly excites
- L27:793 weakly suppresses
- L26:1761 weakly excites
- L26:2136 weakly excites
- L27:2741 weakly suppresses
- L27:934 weakly excites
- L25:2875 weakly excites
- L25:597 weakly excites
- L25:2341 weakly suppresses
- L27:1727 weakly excites
- L27:1572 weakly suppresses
- L26:1077 weakly suppresses
- L25:2837 weakly suppresses
- L27:459 weakly excites
- L25:2214 weakly suppresses
- L25:87 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
- time
- 时间
- time
- Time
- Time
- _time
- 時間
- -time
- (time
- 时间和
Suppresses
- iego
- ató
- isting
- ilian
- Matchers
- стро
- eldorf
- @update
- cede
- atas