"latter" and "former" referring back
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. 6 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 "latter" (or "former") when it points back to the second (or first) of two things just mentioned. Turning it up made the model's text drift into loose, rambling story-like sentences, which doesn't clearly relate to the trigger.
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.
-
: should he send the boys off to jail forthwith, or should he see whether they might not be able to straighten themselves out? He decided on the latter course, and in looking for a means of rehabilitation he selected an unusual instrument:
-
for crop yields by all four methods. Overall, global yields increase ~10% under the three SGs and decrease 5% under emissions reduction, the latter primarily due to reduced CO2 fertilization, relative to business as usual by the
-
primarily as a nutritive process2. Health and illness, in this commonsensical view, are functions respectively of optimum nutrition and the lack of it. The latter could have two forms – over nutrition and under nutrition. This two-fold classification of
-
plumage with a black back.⏎- ‘Neophytes at duck identification often confuse a bufflehead with a male hooded merganser when the latter's crest is raised showing a large patch of white feathers.’⏎- ‘These
-
opportunities may also be associated with the use of carrier multiplication in solar-fuel technologies and specifically, the production of hydrogen by photo-catalytic water splitting. The latter process requires four electrons per water molecule and its efficiency can be dramatically enhanced if these
-
the expansive power of the State and society over individuals? Not always successful, and fraught with problems, but relevant enough for an intellectual tradition to be discovered. The latter can be defended, but the former demands us to jettison its nationalistic
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 need to get this information to the students. This is a great way to show what you really want to learn and share what you care about the students, and the final presentation to the students of the class, so you can use the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Top, and William poured volleys of buckshot upon the enemy, then charged them with his bayonet. He recollected that one of the last shots of Gettysburg was fired at him.⏎“Next morning
-
Picchu, the mountain behind Machu Picchu. Alternatively, climb to the top of the less crowded Machu Picchu peak, which catches the first rays of the rising sun. Both mountains overlook the entire site, down to the river below
-
.⏎Pluto is the last one of the major planet. It's definitely different from the others. Discovered in
-
files, among them first Pdf files consists of Textbook Questions Answers whereas in second Pdf files you will get the additional extra questions for practices. In the third and last Pdf files, you will get full exercise of this chapter.⏎You
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
- L3:2400 excites
- L5:110 suppresses
- L4:2665 weakly excites
- L5:1698 weakly excites
- L0:1672 weakly suppresses
- L5:169 weakly excites
- L0:60 weakly excites
- L5:429 weakly suppresses
- L5:50 weakly excites
- L1:2340 weakly suppresses
- L0:2877 weakly suppresses
- L5:1891 weakly suppresses
- L5:631 weakly excites
- L4:1128 weakly suppresses
- L4:867 weakly suppresses
- L1:867 weakly excites
- L3:1340 weakly excites
- L3:1772 weakly excites
- L2:70 weakly excites
- L5:1782 weakly excites
- L4:850 weakly suppresses
- L4:2808 weakly excites
- L5:1195 weakly excites
- L0:1182 weakly excites
Feeds later layers
- L7:902 excites
- L9:1101 weakly excites
- L7:559 weakly excites
- L13:2489 weakly suppresses
- L8:2806 weakly excites
- L8:858 weakly excites
- L19:1260 weakly excites
- L8:901 weakly excites
- L8:1542 weakly suppresses
- L19:154 weakly excites
- L25:522 weakly suppresses
- L21:2754 weakly suppresses
- L8:1369 weakly suppresses
- L19:2058 weakly suppresses
- L13:1286 weakly suppresses
- L8:2636 weakly excites
- L13:1040 weakly excites
- L10:2980 weakly suppresses
- L16:2091 weakly suppresses
- L12:2906 weakly suppresses
- L8:1563 weakly excites
- L17:2322 weakly suppresses
- L14:2124 weakly suppresses
- L12:2576 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
- Yourself
- matrimon
- ye
- yourself
- [max
- yyyy
- -scrollbar
- yy
- INESS
- tamil
Suppresses
- inos
- 利亚
- ulos
- 能看到
- aina
- entials
- %",
- 但他们
- voir
- %",⏎