'of' after made up, composed, consists
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. 8 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 'of' (or 'from') right after a verb that describes what something is made of, such as 'made up of', 'composed of', 'consists of' or 'comprised of'. It seems to expect a list of parts next. Turning it up made the text ramble without a clear link to materials.
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.
-
old on Friday April 5th in 1974. On what day⏎of the week was she born?⏎A tetromino is made up of four squares joined edge to edge. Can⏎this tetrom<|endoftext|>Congress, C
-
an oil reserve the size of England⏎The tar sands are a source of oil buried below the Boreal Forest of northern Alberta. The tar sands are made up mostly of sand. Only 10-12
-
illnesses approximated towards itself another Vedic idea of evident practical value – that everything in the universe is composed primarily of agni (fire – the heating principle) and soma (moon – the cooling
-
one was adopted by the tribal elders. It consists of seven stripes, the central being 1/6 the size of the other stripes. The hoist consists of a triangle edged in black with an semi-circle of red over black, separated by
-
was able to maintain itself near sea level by growing upwards. That reef is now over 516 feet (160 m) thick and comprised of mostly post-Miocene limestone with a layer of upper Miocene (Tertiary g
-
not dehydrate during the roasting process, has a very small shrinkage, and has relatively stable physical and chemical properties.⏎The slurry used is mainly composed of various components of water suspension colloid. When making embryos, the slurry can
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 have a 90% chance of success on my first attempt at the exam for the exam, and 90% on the second attempt. I have a 90% chance of success on the second attempt. What is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
devices(2427) Terms.⏎The official website of the Tokyo 2020 - Summer Olympic Games - 3x3 event 2021 - FIBA.basketball⏎Here to
-
functions is vital. Some of the essential facts you will need to note includes:⏎- Every function should have a unique name⏎- A function name must be followed by parentheses⏎- Functions need to equipped with return type⏎- A function body has
-
launched by Expeditions, an independent network of scholars in the human sciences. For 11 years, we have been holding in Gozo (Malta) a summer school for anthropologists and social scientists, focused on the practice of fieldwork
-
letters: A D E L M N O P S.⏎No direct anagrams for endoplasm found in this word list.⏎Words formed by adding one letter before or after endoplasm (in bold), or to adelmnops in any
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
- L4:2702 weakly excites
- L4:2991 weakly excites
- L7:2688 weakly suppresses
- L6:13 weakly excites
- L2:2825 weakly excites
- L1:2054 weakly excites
- L7:500 weakly excites
- L6:778 weakly excites
- L7:2456 weakly suppresses
- L7:2590 weakly excites
- L7:2474 weakly suppresses
- L1:2079 weakly excites
- L6:2836 weakly excites
- L6:12 weakly excites
- L2:1874 weakly suppresses
- L3:2379 weakly excites
- L0:1702 weakly suppresses
- L6:1139 weakly suppresses
- L7:196 weakly suppresses
- L6:2377 weakly excites
- L4:2111 weakly excites
- L5:812 weakly suppresses
- L5:1671 weakly excites
- L5:470 weakly excites
Feeds later layers
- L9:1595 excites
- L12:1544 weakly excites
- L10:1539 weakly excites
- L9:796 weakly excites
- L17:2888 weakly excites
- L10:2120 weakly suppresses
- L16:2068 weakly excites
- L14:2374 weakly excites
- L10:1406 weakly excites
- L24:2111 weakly excites
- L12:1531 weakly suppresses
- L15:1836 weakly excites
- L12:1636 weakly excites
- L23:638 weakly excites
- L11:312 weakly excites
- L9:2132 weakly excites
- L11:600 weakly suppresses
- L18:2387 weakly suppresses
- L22:2710 weakly excites
- L12:1859 weakly excites
- L24:1048 weakly suppresses
- L9:881 weakly excites
- L10:2300 weakly excites
- L9:2028 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
- comprised
- volunteers
- consisted
- 组成
- getContent
- naires
- composed
- entimes
- acquainted
- contained
Suppresses
- versions
- .Put
- 方向
- Alternate
- redirect
- ender
- american
- Directions
- offs
- secure