period after abbreviation in citations
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. 5 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 reacts to the dot that follows a short abbreviation in a reference, like "Vol.", "Oct." or "var.", where a number or name is expected next. Turning it up made the model's text drift into unrelated topics, so the steering results do 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.
-
to all use restrictions contained in the Terms and Conditions of Use of the BioOne website.⏎Vol. 65 • No. 2<|endoftext|>FC: Eugene V. Debs
-
A central point; a point of concentration. 2⏎v. t. - To bring to a focus; to focalize; as, to focus a
-
london-history-1216214947983326-8 по слайдам⏎2008 by Mag. Andreas Hofer and the 4 DHWIM⏎It is thought that London
-
fractures, not less, in women because women have less bone<|endoftext|>WebMD Medical News⏎Daniel J. DeNoon⏎Louise Chang, MD⏎Oct. 9, 2012 – About 13,00
-
: Three varieties of this North American subspecies occur in BC. Fruits black or purplish-black; nutlets slightly wrinkled or pebbly var. melan
-
_html/wp-content/plugins/ultimate-google-analytics/ultimate_ga.php on line 524⏎GUJHS. 2007 March; Vol. 4, No. 1⏎Sophie Cowan, ‘0
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 problem is to find the value of this expression, and the answer is something. So, the problem is to find the value of this expression, and the answer is something. This is a problem, but it's not.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
40 percent of excess body weight. However, according to a new study, 21 percent of gastric bypass surgery patients develop a drinking problem.⏎Bariatric (weight loss) surgery has become popular among Americans as an easy weight loss solution
-
a point that Dr. Pretlow called “alarming.” But why? He noted the tendency, in some quarters, to jump to a solution like bariatric surgery, and rhapsodize about what an acceptable safe effective treatment alternative it
-
aparoscopic Gastric Bypass ( A combined restrictive and malabsorptive procedure)⏎With laparoscopic techniques, the bariatric surgeons staple across the top portion of the stomach to create a very small stomach pouch
-
patients were four times more likely to say they had received treatment for substance abuse when they hadn’t.⏎“This indicates that treatment programs are underutilized by bariatric surgery patients with alcohol
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
- L2:1769 suppresses
- L3:699 suppresses
- L1:1728 excites
- L3:1161 suppresses first digit after 'more than'
- L1:1135 weakly suppresses
- L0:456 weakly suppresses
- L2:2883 weakly excites
- L3:51 weakly excites
- L3:1753 weakly excites
- L1:1439 weakly suppresses
- L0:1757 weakly excites
- L2:825 weakly excites
- L2:2495 weakly excites
- L0:847 weakly suppresses
- L0:2484 weakly excites
- L4:1013 weakly suppresses
- L4:1218 weakly suppresses
- L2:1005 weakly suppresses the '6' in years like 196x
- L4:869 weakly excites
- L1:2798 weakly excites
- L3:1915 weakly suppresses
- L2:868 weakly excites line breaks inside lists
- L4:2005 weakly suppresses
- L1:947 weakly suppresses
Feeds later layers
- L8:517 excites
- L13:2803 excites
- L7:765 excites
- L12:792 excites
- L9:897 weakly excites
- L15:998 weakly excites period inside an abbreviation
- L7:887 weakly suppresses
- L15:1976 weakly excites
- L9:2898 weakly excites
- L23:505 weakly suppresses
- L19:1337 weakly excites
- L16:2128 weakly excites
- L26:1888 weakly excites
- L6:82 weakly excites
- L7:346 weakly excites
- L6:1478 weakly excites
- L16:1319 weakly excites
- L14:2798 weakly excites
- L26:1391 weakly excites
- L10:62 weakly suppresses
- L7:2898 weakly excites
- L15:1117 weakly excites
- L13:2513 weakly excites
- L26:1599 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
- 护肤品
- /legal
- intest
- /books
- 进出口
- 聯絡
- 肢
- .Buttons
- .’”⏎⏎
- ventus
Suppresses
- anvas
- plode
- rage
- झ
- زال
- -space
- ingly
- <algorithm
- oldown
- unce