the 9 in years starting 19
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. 23 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 second digit of years that begin with 19, such as the 9 in 1950s or 1971. Turning it up did not change the topic of what the model wrote, so its effect on the text is hard to see.
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.
-
monaut. Flew on Soyuz 7, Soyuz 16. More...⏎Grechko Grechko, Georgi Mikhailovich (1931-) Russian engineer cosmonaut. Flew on Soyuz 1
-
atta Facts⏎Mohammad Hatta (1902-1980), one of the foremost intellectuals in the Afro
-
employment rates for African Americans have been far higher than those of whites for the past 50 years, even in good times. In fact, since 1960 the black unemployment rate has been about twice the white rate. Had blacks
-
was to build a device to read the traffic tapes directly and eliminate the tedious manual work. The Intel 8008 microprocessor was announced in 19
-
976 at 16.8 cigarettes a day.⏎20% of men currently smoke – the lowest value since the series began in the 1970s. In 1974, half (51%)
-
8 election to hard-line nationalists who institutionalised apartheid. He continued to work for reconciliation and emphasised the British Commonwealth's positive role until his death in 19
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: how to use a Ruffle Sware in a Mufsa Muzari. Mufa Muzari is an auster and auster, and the Muzari is a part of the Tariq and Muf
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
22:6⏎“Those things, which ye have both learned, and received, and heard, and seen in me, do: and the God of peace will be with you”
-
479 BC), a Chinese scholar, politician and philosopher who lived in the Spring and Autumn period. The main body of the Confucian canon comprises the Four Books and the Five Classics (四書五經), texts which have been traditionally
-
2008 IDE⏎- Create an empty project by select File >> New >> Project. Select General >> Empty Project and type in the project name.⏎- Under folder Source, right-click Select Add >> New Item.⏎- Under Code, Select
-
the water. If the water stays cold then you could keep the chicken there a long time, however, once the water is above 40 F (4.4 C) then limit the time.⏎Should I soak chicken in salt water?⏎
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
- L21:153 suppresses
- L22:1855 excites
- L19:417 suppresses
- L21:582 excites
- L22:399 excites
- L20:2181 suppresses
- L18:2361 excites
- L21:2003 suppresses
- L21:475 suppresses
- L22:1005 suppresses
- L18:2926 weakly suppresses
- L2:1321 weakly excites
- L17:1336 weakly excites
- L16:1036 weakly suppresses
- L18:569 weakly suppresses
- L10:482 weakly excites
- L19:2605 weakly suppresses
- L21:1329 weakly suppresses
- L20:355 weakly excites
- L21:295 weakly excites
- L18:576 weakly excites
- L17:744 weakly excites
- L17:1037 weakly suppresses
- L18:2656 weakly suppresses
Feeds later layers
- L26:1009 suppresses
- L26:745 suppresses
- L24:747 suppresses
- L27:331 suppresses
- L24:340 suppresses
- L25:735 excites
- L27:870 weakly suppresses
- L27:1107 weakly suppresses
- L26:132 weakly suppresses
- L24:174 weakly excites
- L24:105 weakly suppresses
- L26:116 weakly suppresses
- L24:512 weakly excites 'the' inside organisation and convention names
- L25:1623 weakly excites
- L25:414 weakly suppresses
- L27:1347 weakly excites
- L24:2460 weakly suppresses
- L25:812 weakly excites
- L26:934 weakly suppresses
- L27:2059 weakly excites
- L27:851 weakly suppresses first names of famous people
- L24:27 weakly suppresses
- L27:470 weakly suppresses
- L25:1487 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
- BJP
- CBD
- IClient
- apist
- SPDX
- Jedi
- WWE
- HuffPost
- iPod
- opause
Suppresses
- 萨
- チェ
- 佩
- 迈
- 查
- Mon
- 达
- Char
- Chip
- 芒