the word 'still'
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. 20 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 'still', as in 'still reflects' or 'still survive today', meaning 'continuing to be'. When it was turned up, the model kept inserting the word 'still' over and over, so the link is clear.
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.
-
, faculty, postdoctoral scholars, administrators, organizations) served.⏎John Locke ‘s work Some Thoughts Regarding Education was written in 1693 and still reflects conventional education priorities in the Western globe. Every youngster with disability will be identified
-
Canada. The work done by the American Bureau was monumental, well informed and incorporated the most advanced scholarship available at the time. In many respects, the information is still useful today, although prudence should be exercised and the reader should consult some of
-
is so rare for a lot of people — made him someone who became a touchstone for folks' lives. And that's why they responded to him, and still do.⏎"Keep Ya Head Up" [is] a song that is really
-
land so important?⏎- What are the strengths of the Heartland Theory?⏎- What are the weaknesses of the Heartland Theory?⏎- Does the Heartland Theory still apply today?⏎- Key points to remember about Mackinder’s Heartland Theory⏎
-
trim- a post he held until 1860. Charles Lanyon firmly made his mark on Victorian Belfast’s urban landscape and many of his works still survive today. Some of his notable works include Queen’s College (184
-
80 churches it destroyed was Saint Paul’s Cathedral. The Gothic church was rebuilt by Sir Christopher Wren in Renaissance style and it received its famous dome, which still is, despite fierce competition from the London Eye, one of the most popular viewpoints
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: what if the people who have the most people in the wrong of their people have the most people in the wrong of their wrong ones, and the people who have the most people in the wrong of their wrong ones have the most people in the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
)o/a|⏎|Beam:||33 ft (10.06 m)|⏎|Draught:||11.5 ft (3.51 m)|⏎|Speed:||16 knots (29.
-
Highest point||top of flag staff|⏎|- elevation||15.75 ft (4.8 m) over stone base|⏎July 3, 19
-
and had seven locks, each 30 m (98 ft) long, 6 m (20 ft) wide and 1.5 m (4.9 ft) deep. It officially opened in 1825
-
an asteroid about 5-10 meters (15-30 feet) in diameter exploding in the air between 15 and 20 km (nine to 12 miles) above sea level. Nobody was injured as a
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
- L17:2306 excites
- L16:2352 suppresses
- L19:2487 excites
- L17:2393 suppresses
- L16:1470 excites
- L16:2301 suppresses
- L17:709 excites
- L13:2541 excites
- L13:2890 excites "still" or "as long as" meaning continuing
- L19:2254 suppresses
- L10:1848 suppresses
- L16:2123 excites
- L17:2618 suppresses
- L14:2829 weakly excites
- L5:2053 weakly suppresses
- L18:3044 weakly suppresses
- L13:2153 weakly excites
- L9:1699 weakly suppresses
- L11:2401 weakly excites
- L19:1690 weakly excites
- L14:1302 weakly excites
- L12:2351 weakly excites
- L13:1909 weakly suppresses
- L8:2072 weakly suppresses
Feeds later layers
- L23:1638 strongly excites
- L22:2054 suppresses
- L26:2595 excites
- L27:1280 excites
- L22:1795 suppresses
- L24:2256 excites
- L25:2237 excites
- L21:724 weakly excites
- L23:2859 weakly excites
- L22:1220 weakly excites
- L27:2346 weakly excites
- L23:525 weakly excites
- L22:891 weakly excites
- L23:988 weakly suppresses
- L21:1772 weakly excites
- L26:1243 weakly excites
- L22:1432 weakly excites
- L22:2508 weakly excites
- L22:1774 weakly excites
- L24:2111 weakly suppresses
- L23:2817 weakly excites the word "no" in "no longer"
- L24:1746 weakly excites
- L22:2528 weakly excites
- L21:2295 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
- 仍然
- 依然
- still
- 仍是
- still
- 仍在
- 仍然是
- 仍
- 依旧
- 仍有
Suppresses
- tab
- 已久
- ComVisible
- inks
- аниз
- 注
- сть
- づ
- atted
- buổi