'the more X' clause before 'the more'
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. 19 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 reacts to the comma that ends a phrase like 'the more you drink, the more...', where the model is about to write the matching 'the more' half. When it was turned up, the model fell into a loop of 'the more the more the better'.
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.
-
this is done, the company loses vital manufacturing and quality information. This information accurately collected is the only basis of implementing quality improvement. The more granular the information, the more effective it can become.⏎If the only options are in-tolerance and
-
are being used to validate this identity combination, and to prove a virtual person is indeed the same person in real life. The more value an identity has access to, more effort and work is required to validate and establish secure authentications for it.⏎
-
2 studies, found there was a dose-response association between intake of plant protein and risk of death (from any cause) - the more plant protein in the diet, the lower the risk of death.⏎What foods are high in plant protein? Leg
-
the file names. ACTOZ is the name we shall use throughout this user's manual.⏎Actoz is a very powerful instrument. The more data it contains, the greater its potential for thorough analysis and unexpected results. One of the greatest qualities
-
few (or 30) seconds or your water’s a couple degrees off temperature, you’re NOT going to ruin your film. The more accurate you are,
-
stress.⏎Stress is an underlying factor in illness as it reduces your immune system’s ability to fight off antigens. This means that the more stressed you are, the likelier you are to pick up infections and fight illness.⏎Therefore, a
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: If I can get back to old friends in 2020, then maybe it would help to keep working at school now, by working as much as possible, but instead of working for money, working to save money in working.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a fraction of a second.⏎The discovery has been accepted for publication in Physical Review Letters.⏎Gravitational waves are ripples in the universe caused by some of the most energetic cosmic events, from exploding stars to supermassive black hole mergers
-
two end systems⏎B. It specifies the quality of the data & connection⏎C. It specify the size of data⏎D. All of the above⏎6-4. The class-based addressing is also known as⏎A
-
a unique continental and leading agricultural product. One of the countries on the equator line benefits from its climate and agriculture production. Before you
-
research.⏎LaMantia and his colleagues looked at brain tissue from mice missing the same region. They found that the mice have unusually few neurons in two of the six layers of the cerebral cortex — the outer portion of the brain. These neurons
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
- L18:2316 excites
- L17:30 suppresses
- L4:52 excites
- L16:2248 suppresses
- L18:2979 excites
- L9:24 excites "the" before a named organisation or title
- L16:419 excites
- L18:641 suppresses
- L12:435 suppresses
- L15:1848 weakly suppresses
- L18:2308 weakly suppresses
- L12:3007 weakly suppresses
- L2:1473 weakly excites
- L18:239 weakly suppresses
- L11:1168 weakly excites
- L8:636 weakly suppresses
- L5:125 weakly excites "of" after all three, both, ten
- L6:426 weakly suppresses
- L18:602 weakly suppresses
- L4:73 weakly suppresses
- L4:2719 weakly suppresses
- L10:1377 weakly suppresses
- L7:63 weakly excites
- L3:59 weakly excites
Feeds later layers
- L26:507 strongly excites
- L27:127 strongly suppresses
- L27:399 strongly excites
- L22:171 strongly excites
- L24:49 strongly suppresses
- L21:863 excites
- L21:310 excites
- L21:2949 excites comma closing 'the more X, the more Y' clauses
- L27:121 suppresses
- L20:64 suppresses
- L21:758 excites
- L25:247 excites
- L22:197 suppresses
- L22:353 suppresses
- L23:94 suppresses
- L20:593 excites
- L20:2278 excites
- L20:738 excites
- L24:92 suppresses
- L20:2321 excites
- L25:1796 excites
- L22:360 weakly excites
- L26:693 weakly excites
- L22:1589 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
- the
- the
- →the
- _the
- The
- .the
- The
- THE
- thed
- …the
Suppresses
- RIEND
- awning
- REEN
- ��
- ày
- reen
- 余人
- Bosch
- emics
- uplicates