space between month and day number
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. 18 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 gap between a month name and the day of the month, as in 'October 31' or 'September 26'. Turning it up or down did not produce anything obviously related to dates, so the effect on the model's writing 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.
-
is part of a series of features highlighting freshman seminars offered at Princeton this semester. Other features:⏎Exploring the science and nuance of facial perception⏎Posted October 31, 2011; 12:00 p
-
-economic status, ethnicity, and religion?⏎- Are they conscious of themselves as having a “self?”⏎Submit your assignment to the Discussion Area by Saturday, November 22, 2014.<|endoftext|>Latest posts by Jessica Edwards (
-
Seehofer has blamed Merkel's refugee policy for the AfD's success.<|endoftext|>Posted by Jayme Blaschke⏎University News Service⏎September 26, 2011⏎Victor Frankenstein’s infamous monster
-
Annap. Harb. Ctr.: 410/266-1447.⏎Tiny Tots Fall for Nature Tues. Oct 12 (10:30am-noon) Three- to five
-
, plays, poetry and non-fiction texts.⏎All our teachers are English specialists.⏎For more information contact the head of department, Miss S Beadle.⏎In Years 10 and 11, all pupils study for the AQA GCSE
-
notes reflecting on the moment. Listen to learn what inspired the project and hear some of the writing that hikers left behind.⏎Current Event October 20, 2017⏎Middle and high school students can spend
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: is the * ( or ) and the the- is, in the is- or - to- the, or - is to, or the- is, or - to, or is, or is-
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, Serbia, Singapore, Spain, Sri Lanka, Sudan, Tajikistan, Tunisia, Turkey, Ukraine, Yemen⏎The Wiki is a web-based, multi-lingual and interactive project - this means we need your input to ensure the pages
-
16/18 vaccine may prevent cervical cancer just as effectively as three doses (Kreimer et al., 2011). A one- or two-dose program could lower the cost of vaccination and encourage the global dissemination
-
]⏎The World As It Could Be Curriculum and Resource Guide for Teachers & Organization Leaders is now available through the Education Resources Information Center (ERIC) at http://www.eric.ed.gov/ERICWebPortal/detail?accno=ED532
-
is the most common example of a 'drying oil' but rapeseed, cottonseed, peanut, corn oil and safflower oil may also self-heat. Fish oils are notorious for their self-heating properties. Mineral oils like
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
- L13:923 suppresses word right before a number or year
- L4:53 excites blank slot before a number
- L6:715 excites
- L10:220 suppresses hyphen inside compound words
- L14:140 suppresses
- L16:1407 excites
- L9:7 excites
- L7:193 suppresses space before day after month
- L2:404 excites
- L1:217 suppresses
- L3:421 suppresses
- L14:348 suppresses
- L4:185 weakly excites
- L16:396 weakly suppresses
- L6:17 weakly suppresses
- L3:342 weakly excites
- L4:490 weakly suppresses decade digit in years like 1930s
- L1:1609 weakly excites
- L4:628 weakly suppresses
- L8:350 weakly suppresses blank space before a year
- L5:97 weakly suppresses
- L0:1301 weakly suppresses
- L2:75 weakly suppresses
- L17:426 weakly suppresses
Feeds later layers
- L22:2384 suppresses
- L20:163 suppresses
- L23:239 suppresses
- L25:422 suppresses
- L21:302 suppresses
- L25:304 suppresses
- L27:553 suppresses
- L27:317 suppresses
- L27:981 suppresses
- L26:418 weakly suppresses
- L21:29 weakly suppresses
- L27:527 weakly suppresses
- L20:1779 weakly suppresses
- L26:526 weakly suppresses
- L21:1358 weakly excites
- L27:431 weakly suppresses
- L22:265 weakly suppresses
- L20:1067 weakly suppresses
- L22:372 weakly suppresses empty spot before a number or "3D"
- L26:459 weakly suppresses opening bracket before an acronym
- L27:344 weakly suppresses
- L25:513 weakly suppresses
- L27:2114 weakly excites
- L27:220 weakly suppresses end-of-text token between documents
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
- казан
- name
- Leap
- Pract
- practice
- innings
- nak
- Practice
- ứng
- Lent
Suppresses
- individual
- ǧ
- automatic
- →→⏎→→⏎→→⏎
- ugging
- aggregate
- uitar
- utar
- auty
- ipelines