numbers right after 'these' or 'first'
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 fires on the word just before a number, especially after words like 'these', 'first', or 'those same', as in 'these 120 films'. When turned up, the model's text drifted into rambling, repeated phrases about being honest or about the news, with no clear theme.
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.
-
with closed-captioning. Users will also find 164 film clips (80 in English; 84 in French in the French part).⏎These 120 films are only a fraction of NFB productions on Canada’s ethnic
-
food. The various states submitted their top 10 plants to the association, and from that a list was produced for the Southeast region of the country.⏎These 10 browse species are:⏎American beautyberry (French mulberry)⏎Beg
-
and medicines.⏎By planting 20 million trees, the earth and its people will be provided with 260 million more tons of oxygen. Those same 20 million trees will remove 10 million tons of CO2
-
only 5 frogs and⏎4 worms, also to pick out at least one of each you would need to⏎pick out 15 sweets because the first 9 could be mice and the next 5⏎you pick out could be frogs
-
the conclusions of the detailed comparison analysis highlight several elements that will allow the team to refine their method and produce even more detailed maps.⏎The valuable information derived from these 2 projects forms a solid basis to support policy initiatives in Flanders as well as
-
.⏎The majority of patients (85%) fail to clear the virus within 6 months and develop chronic hepatitis C. These patients are relatively well in the first 2 decades after acquiring the infection. However in 20% of these carriers
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 am a teacher in a class. I have a class with 3 students. There are 4 students in total. I have a class with 3 students. There are 4 students in total. I have a class with
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of effectiveness) each year produces over $350,000 more income for his/her students compared to an average teacher. On the the other hand, a poor teacher (one in the bottom quarter in terms of effectiveness
-
Rosey the Riveter, the woman, sporting a red bandana, her bicep in the air, saying "...We Can Do It!". Or the the poster warning against spies with "...Loose Lips Sink Ships".⏎By the time you
-
, as well as nationality, were increasingly reduced to one dimension as the climate of insecurity turned nationalism into a state religion.” The all-encompassing potential for abuse within nationalism can redefine and control a people with brutal
-
to provide you, the budding programmer, the the knowledge you need on programming in an easy and fun manner. In this course I
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
- L1:1874 excites
- L0:178 weakly suppresses
- L1:740 weakly suppresses
- L1:2341 weakly excites
- L0:2049 weakly excites
- L1:2579 weakly suppresses
- L0:2424 weakly suppresses
- L1:331 weakly suppresses
- L0:2989 weakly suppresses
- L0:193 weakly excites
- L1:1099 weakly excites
- L0:1224 weakly suppresses
- L1:2683 weakly excites
- L1:1007 weakly suppresses name pieces inside a multi-part word
- L1:604 weakly excites
- L1:1660 weakly suppresses
- L1:200 weakly suppresses
- L1:693 weakly suppresses
- L0:2495 weakly excites
- L1:124 weakly suppresses
- L1:2032 weakly excites
- L1:2593 weakly suppresses
- L1:1492 weakly excites
- L1:2569 weakly suppresses
Feeds later layers
- L9:537 weakly suppresses
- L3:1138 weakly suppresses
- L17:2970 weakly excites
- L4:159 weakly excites
- L7:2915 weakly suppresses
- L25:1435 weakly suppresses
- L24:1086 weakly suppresses
- L20:100 weakly excites
- L18:1547 weakly excites
- L23:1316 weakly suppresses
- L14:1020 weakly excites
- L23:105 weakly excites
- L13:2729 weakly excites
- L3:1538 weakly excites 'what' as a free relative
- L9:376 weakly suppresses
- L5:1194 weakly suppresses
- L16:1433 weakly suppresses
- L16:2856 weakly suppresses
- L18:1943 weakly suppresses
- L15:2762 weakly suppresses
- L24:303 weakly suppresses
- L6:2145 weakly excites
- L4:1851 weakly suppresses
- L5:2409 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
- "..
- insured
- dots
- points
- 剂
- notation
- etes
- positions
- atorial
- ickets
Suppresses
- structuring
- 加载
- 实现
- reeze
- yên
- 另一个
- 挖掘
- 还将
- магазин
- 兼顾