the letter t starting a word fragment
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. 27 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 responds to a lone 't' that begins the first piece of a longer word, as in 'Settlor', 'trollhavre' or 'tubes'. Turning it up made the writing repeat words and phrases, so the effect is not easy to read from the output.
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.
-
of two categories – testamentary or living trusts. A testamentary trust is activated by a provision in the Settlor’s Will at the time of death. A living trust activates once all formal
-
It can be at any time and for any reason. However, if the trust is an irrevocable living trust the Settlor does not have the authority to terminate the trust after it is established.⏎The
-
has spread [from Denmark] to Norway and Sweden, although the Lokke-name has been abandoned and the plants are called “trollhavre” (troll oat) and “pukhagre”.⏎On the other hand
-
how a living trust terminates.⏎What Is a Living Trust?⏎The person who creates a trust goes by many names. They can be referred to as the “Settlor”, “Trustor” or “Grantor.” The Settlor appoint
-
N=23) attending music-oriented and regular fourth grades were tested. The stimuli were two semisynthetic pseudo words /ty:ti/ and /tʉ:ti/ with the native vowel /y/ and the non-native vowel
-
valuable of gems, known only to kings.” Pliny described six varieties, among which the Indian, having six pointed angles, and also resembling two pyramids (turbines, whip-tops) placed base to
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:要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要要
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
:⏎- Vitamin E (α-tocopherol)⏎- Causes and Risks for Vitamin E Deficiency⏎-
-
mechanism of the body and when you get injured it turns on to try and protect the body from any further damage. In this process the brain will shutdown/activate/tighten any muscle it needs to, to prevent damage to the frame of the
-
to tropical storm status, with maximum sustained winds of 65 mph (105 km/h).⏎On September 14, Humberto had reached post-tropical cyclone status, with winds at 40 mph (65 km
-
so by 6 months."⏎A study published in the July 27, 2009 online version of Pediatrics involved establishing and evaluating the Not-On-Tobacco (N-O-T) program in 8 Maryland high schools. The
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
- L22:316 strongly excites
- L23:1407 strongly suppresses
- L20:1040 suppresses
- L24:214 excites
- L19:529 excites
- L21:2453 suppresses
- L23:2170 suppresses
- L22:2148 excites
- L24:1220 excites
- L21:2602 suppresses
- L21:761 suppresses
- L21:2711 excites
- L26:1564 excites
- L22:1502 excites
- L16:76 excites
- L8:7 excites
- L20:1815 suppresses
- L26:2170 suppresses
- L8:14 excites
- L23:3034 excites
- L21:2617 suppresses
- L8:10 excites
- L1:21 excites
- L18:161 excites
Feeds later layers
None: this is the last layer.
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
- 一定程度
- culus
- Nature
- gem
- q
- Nature
- ulture
- LOOR
- j
- nature
Suppresses
- 要
- 都要
- 也要
- 要在
- 又要
- 还要
- 就要
- 我们要
- 要加强
- 要用