"well" in "as well as"
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. 15 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 lights up on the word "well" in the phrase "as well as", where something is being added next to something else. Turning it up or down did not make the model write about that phrase, so the link to the output is unclear.
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.
-
circuits that can be sprayed onto anything using modified inkjet technologies - no expensive clean rooms necessary. That has the attention of dozens of fledgling businesses, as well as industry veterans Dow, IBM, Philips, and Xerox, which are pouring
-
ships so that they could beat competitors and outwit planetary security forces. Almost every smugglers vessel had improved light speed and sub-light speed drives for increased speed as well as boosted weapons systems to fight their way out of tough situations. Another of the
-
types of restoration strategies will be used for this project. At least 50,000 native trees and shrubs will be planted during the project as well as several thousand wildflowers and grasses in order to restore approximately 25
-
, including interviews and reporting on the maker market and the wider technology world. Our standards page is intended as a guide to best practices that Adafruit uses, as well as an outline of the ethical standards Adafruit aspires to. While Adafruit
-
concessions. If history tells us anything, his turn toward a more populist brand of politics has more to do with the energy of reform movements within Honduras itself, as well as throughout the rest of Latin America, than any personal awakening.⏎From the shift
-
with the precise soundbites that will accompany them which match what was actually recorded in interviews⏎Let’s jump in and examine each of these in detail, as well as learn how to write each one.⏎What is a documentary film treatment?⏎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: I AM A TALKING FISH, AND I AM A WALKING FISH, AND I AM A FISH BUT I AM A FISH, AND I AM A FISH, AND I AM A FISH, AND I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
9/10)k-2, where k is the number of digits of 2n. The number of digits can be estimated as n log102. Thus, the probability is about cxn, where c = (1
-
Servicing Ratio is a measure used by lenders to assess a borrower's ability to repay their debts. The DSR is calculated by dividing a borrower's monthly income by their monthly debt repayments, including mortgage payments, credit card payments, and other
-
.⏎Hence, the given lines are parallel.⏎Let's consider another set of equations.⏎y = 2x +3 and 4y = –2x+10⏎On comparing y = 2x +3 and 4
-
to identify parallel and perpendicular lines.⏎Let's take some examples to understand this.⏎Considering two equations:⏎y = 2x +3 and y = 2x+5⏎On comparing y = 2x +3 and y =
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
- L12:2659 suppresses
- L14:2732 excites
- L13:1312 suppresses
- L14:2020 suppresses
- L12:2696 suppresses
- L14:2768 suppresses
- L14:2943 weakly suppresses
- L5:108 weakly suppresses
- L9:2939 weakly suppresses
- L12:2058 weakly excites
- L14:986 weakly suppresses
- L11:2336 weakly suppresses
- L8:1069 weakly suppresses
- L14:1361 weakly excites
- L8:711 weakly excites
- L13:781 weakly suppresses
- L10:1205 weakly excites
- L0:850 weakly suppresses
- L14:3004 weakly excites
- L11:1655 weakly suppresses
- L4:1788 weakly suppresses
- L14:2880 weakly suppresses
- L7:660 weakly excites
- L13:2824 weakly excites
Feeds later layers
- L16:1084 suppresses
- L16:473 excites
- L20:1763 suppresses
- L16:3060 suppresses
- L16:2880 suppresses
- L16:2386 suppresses
- L17:2450 weakly suppresses
- L16:2329 weakly suppresses
- L18:2331 weakly suppresses
- L19:2800 weakly suppresses
- L18:1573 weakly excites
- L16:533 weakly excites
- L17:2727 weakly excites
- L21:2925 weakly excites
- L19:2225 weakly suppresses
- L16:774 weakly suppresses
- L16:1956 weakly excites
- L25:2200 weakly excites
- L19:1262 weakly suppresses
- L17:1488 weakly suppresses
- L21:1338 weakly excites
- L16:2790 weakly excites
- L17:632 weakly suppresses unclear: coordinating words in lists
- L24:2510 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
- oft
- 有效的
- chos
- necess
- nas
- 更好的
- monkey
- ammer
- 铤
- estará
Suppresses
- enberg
- 然后
- изма
- tical
- CREMENT
- 并
- 但是我
- IMIZE
- 年以上
- 旋