digits inside years like 2017
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. 13 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 digits inside years, especially the zero in years like 2011, 2017 or 2021. It lights up less on years from long ago, like 48 BCE or 1793. When it was turned up, the model's text became rambling and off topic, so the steering output does not show a clear theme.
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.
-
flu severity, it helps them better understand whether its widespread. An infographic in the article shows just how rampant reports of flu-like symptoms were in late November 2017 as compared to previous years.⏎And while that gives experts plenty of reason
-
11 southern hemisphere influenza seasons.⏎During weeks 29 to 30 (17 July 2011 to 30 July 2011), National Influenza Centres (NICs) from
-
21 - SARS CoV-2: Its Morphology, Genome, and Implications for Therapy and Vaccines⏎September 29, 2021 - Vaccine Types and Their Effectiveness⏎The recent approval of the commercial
-
questions in this exercise. Our subject expert teachers have solved each of these questions in a step by step manner. All these solutions are beneficial for the students of class 10 appearing for the board exams. Students can easily NCERT Solutions for Class
-
, 4 PM⏎AuctionsNov 23, 2021, 1 PM⏎World StampsNov 22, 2021, 7 PM<|endoftext|>At SACRE, the Fluent CFD code
-
electric power generation. Journal of Hydrology (NZ), 34(1), 1-14.⏎Permanent Research Commons link: http://hdl.handle.net/10289/4701⏎Controlled groundwater abstraction
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 text is: "The text is: 1004479a6b0a6e6f0a6e6d479f0a6e6d479f0a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Egypt. Simultaneously, she sought to avenge the death of her son Khemu at the hands of the Order of the Ancients, and in 48 BCE, she killed two of their members, Actaeon and Ktes
-
as a constitutional monarch and appealed for an end to the Revolution. He retired to Grenoble, and was tried for treason and guillotined (1793). His history of the French Revolution, written during his imprisonment, is considered
-
Hidden Ones to resume the fight against the Order, with the former resolving to establish a Roman branch at the expense of her marriage.⏎On 15 March 44 BCE, this culminated in Aya slaying Septimius in personal
-
do wolves like it when it is snowing? We are hoping to answer all of our questions through research.⏎The school bus came at 9:30 on a cold, windy, soggy day and picked us up to
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
- L11:1629 suppresses
- L12:1809 suppresses
- L9:349 suppresses the spot before a year
- L12:1264 excites gap before a year in dates
- L10:1304 suppresses
- L11:1698 suppresses
- L12:940 suppresses
- L11:1246 excites
- L12:2401 excites
- L8:421 excites
- L10:1613 weakly suppresses
- L11:2012 weakly suppresses
- L12:2338 weakly excites
- L11:2739 weakly suppresses
- L11:2213 weakly excites
- L9:415 weakly excites
- L3:2789 weakly suppresses
- L12:2163 weakly excites
- L10:1470 weakly suppresses
- L8:283 weakly suppresses
- L0:1464 weakly excites
- L0:176 weakly suppresses
- L12:2205 weakly suppresses
- L12:3030 weakly suppresses
Feeds later layers
- L17:1051 suppresses
- L14:759 suppresses
- L15:2637 suppresses
- L15:924 excites
- L15:2236 suppresses
- L16:3033 weakly suppresses
- L18:995 weakly suppresses
- L17:52 weakly suppresses
- L15:587 weakly excites
- L15:674 weakly excites
- L17:1241 weakly excites
- L14:555 weakly suppresses
- L15:870 weakly excites
- L18:240 weakly suppresses
- L15:958 weakly suppresses
- L17:45 weakly excites
- L16:2936 weakly excites 'of' after a title or position noun
- L14:1558 weakly excites
- L16:2203 weakly suppresses
- L15:73 weakly suppresses
- L16:1020 weakly suppresses
- L14:443 weakly suppresses
- L15:790 weakly suppresses
- L14:785 weakly suppresses
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
- ProgressHUD
- uder
- тин
- ymous
- obel
- .printStackTrace
- IBE
- ��
- usalem
- gether
Suppresses
- 省
- .generated
- economy
- otos
- "_
- urchases
- "
- środ
- izers
- 犬