"by" introducing a deadline or means
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. 6 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 "by" when it introduces a deadline ("by 2030") or the way something is done ("by venturing into space", "by covering with tar"). Turning it up did not produce a clear change in the writing.
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.
-
at creating new, green jobs and investments in frontline communities bearing the brunt of environmental injustice—they offer a blueprint for both cutting U.S. carbon emissions in half by 2030 and doing so equitably.⏎But neither plan explicitly gr
-
up to seven astronauts.⏎China embarked on its manned space exploration odyssey in 2003 when Yang Liwei, a former fighter pilot, made history by venturing into space aboard the Shenz
-
importance of embracing the natural world.⏎Additionally, [Theme 2] is explored in the poem. Oliver delves into the theme of [Theme 2] by [Provide an analysis of how the theme is portrayed in the poem].⏎Use of
-
to the hub of the vehicle completely, or debris or corrosion may be between the hub and the rotor, causing this runout. A technician will correct runout by completely cleaning the surfaces where the hub and rotor meet. The rotor may also need
-
facilities. Nigeria has pledged its commitment to meeting Sustainable Development Goal (SDG) 6, which is to “ensure access to water and sanitation for all” by 2030. But the National Longitudinal Phone Survey (NL
-
19, he moved<|endoftext|>A team of researchers from the University of California have developed an optical processor designed to offer massive parallelization for deep-learning workloads — by giving a deep-learning workload the job of designing it.⏎"Large-scale linear operations
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: The first time, I have been to a place I have been to a place I have been to a place I have been to a place I have been to a place I have been to a place I have been to a place I have
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
?" Butte said. "It's a simple question that no one had answered in this way."⏎In their paper, published in the current issue of the Public Library of Science ONE, they say there's an established link between Type 1 diabetes
-
brochure is one response to a need identified by people with disabilities. The way we portray people with disabilities and our attitudes toward them are critical to their future...and to ours.⏎Disability vs. Handicap⏎A disability is a condition caused by
-
. The RIFA is widespread, but Florida is also home to a less common species known as the native or tropical fire ant. Neither of the species takes kindly to being stepped on, and they respond to such unwarranted behavior by attacking merciless
-
individuals and their cultures could impede pharmaceutical care. Incongruent beliefs and expectations between the patient and pharmacist could lead to misunderstandings, confusion, and ultimately to drug misadventures. Models and frameworks have been developed that provide descriptions of
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
- L4:858 excites
- L4:1394 weakly excites
- L4:1711 weakly suppresses
- L5:2641 weakly excites unclear: scattered word fragments
- L0:872 weakly excites
- L4:809 weakly suppresses
- L3:381 weakly suppresses 'for' in research institute names
- L4:57 weakly suppresses
- L5:1545 weakly excites
- L4:1460 weakly suppresses
- L4:2893 weakly excites the word 'means' after a subject
- L0:2561 weakly suppresses
- L5:880 weakly excites
- L3:545 weakly excites
- L3:1579 weakly suppresses
- L1:2703 weakly excites
- L2:1477 weakly suppresses
- L5:2392 weakly suppresses
- L2:2018 weakly excites
- L4:486 weakly excites
- L4:1093 weakly excites
- L1:578 weakly excites
- L2:1687 weakly excites
- L4:2363 weakly excites
Feeds later layers
- L8:94 strongly excites
- L8:369 strongly suppresses 'of' after 'learned' or 'because'
- L9:226 strongly excites 'to' starting an infinitive
- L16:1153 excites
- L15:1176 excites
- L14:659 excites
- L8:723 excites
- L18:472 excites
- L7:138 excites 'to' after verbs like prove or time
- L17:472 excites
- L17:1592 excites
- L16:559 excites weak, mixed: verbs and adverbs in claims
- L7:681 excites 'from' after prevent or stop
- L7:15 excites
- L19:385 suppresses
- L19:652 excites
- L15:1234 excites
- L17:187 excites
- L19:617 excites
- L18:715 excites
- L22:200 excites
- L16:1114 excites
- L13:909 excites 'to' before a verb after a noun
- L18:452 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
- 把她
- dig
- 让她
- alte
- イラ
- 对她
- 让消费者
- 把这个
- digitally
- 创造出
Suppresses
- izen
- urance
- concern
- ustr
- consistency
- lemen
- .toList
- 련
- usion
- ounded