"the" in "over the past" time spans
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. 23 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 "the" when it comes right after "over" in phrases like "over the past decade". When it was turned up, the model got stuck repeating "last" and "past", which fits the kinds of words it helps predict next.
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.
-
. “The shellfish results indicate the strength of the growing aquaculture sector.”⏎Supply and demand of farmed oysters have risen exponentially in North Carolina over the past
-
colonies are disappearing fast. According to the nonprofit organization Pollinator Partnership, the United States has lost over 50 percent of its managed honeybee colonies over the past 10 years.⏎Pollinator gardens are just one of the many ways
-
date backwards, using B.C. dates, to the time of the creation of the world. Provisionally, based on the studies published in Biblical Chronology over the past two years, I have come to the year 3932 B
-
by 50% before 2030.⏎This commitment isn’t just good for the planet—it also comes with considerable financial benefits. Numerous studies over the past three decades have proven the value of green building certifications, with a rental premium
-
proximity to our brains, many people are worried that they might be associated with an increased risk of brain tumors. There have been around 30 research studies over the past decade that have looked at this issue and most have found that:⏎- People
-
breakthrough research is Allan Showler, an entomologist in ARS’s Integrated Farming and Natural Resources Research Unit at Weslaco, Texas.⏎“Over the 116 years since the boll
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: though it is a small country, it is a country that is trying to solve its national crisis that is still very serious and unknown. How can we solve this country's problem? annament annament annament annament
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and browse over 110,000 wildlife and science products⏎Multi-currency. Secure worldwide shipping⏎Wildlife, science and conservation since 1985<|endoftext|>A new study recently published showed that for women who are overweight
-
solar winds. After decades of working on that theory, a 1999 study reported evidence that the Sun's magnetic field had strengthened greatly since the 1880s, which brought more attention to the question of whether increased solar
-
problem is relevant for Professor Finnegan's class at WSU.⏎What textbook is this problem found in?⏎Our data indicates that this problem or a close variation was asked in Chemistry: An Atoms First Approach - Zumdahl Atoms 1
-
.11i standard. WPA could be implemented through firmware upgrades on wireless network interface cards designed for WEP that began shipping as far back as 1999. However, since the changes required in the wireless access points (AP
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
- L19:1999 strongly excites
- L19:2563 strongly excites
- L20:1259 strongly excites
- L18:620 strongly suppresses
- L19:2470 strongly suppresses
- L21:1245 strongly excites
- L20:2127 excites
- L19:2536 excites
- L17:1255 suppresses
- L21:2207 excites
- L14:2453 suppresses
- L15:2655 suppresses
- L11:403 suppresses
- L21:1093 excites
- L8:255 suppresses
- L18:2538 suppresses
- L16:1602 excites
- L22:600 excites
- L13:2042 excites
- L20:990 excites
- L15:2575 suppresses
- L12:2964 suppresses
- L17:2719 excites
- L20:2894 excites
Feeds later layers
- L27:1562 strongly excites
- L25:1275 strongly excites
- L26:1094 strongly excites
- L24:1770 excites
- L27:2033 excites
- L24:1194 excites
- L24:2353 excites
- L27:2197 excites
- L24:1794 excites
- L25:2931 excites
- L25:63 weakly suppresses
- L25:2579 weakly suppresses
- L26:2398 weakly excites
- L27:1086 weakly excites
- L27:1335 weakly excites
- L26:1914 weakly excites
- L27:1046 weakly excites
- L26:2514 weakly excites the word 'just' after have or has
- L24:1835 weakly excites
- L27:786 weakly excites
- L26:1525 weakly excites
- L27:1387 weakly excites
- L26:549 weakly excites
- L27:2471 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
- previous
- last
- past
- previous
- 之前的
- last
- Previous
- 之前
- Previous
- (previous
Suppresses
- bcd
- 둘
- unce
- contres
- /cmd
- eries
- 不解
- 轸
- uales
- ->{$