comma after a place name
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. 7 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 responds to commas that separate a city from its country or region, such as 'Sydney, Australia'. Turning it up did not produce any obvious change in the topic or style of 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.
-
, the European Free Trade Association, Chile, Thailand, Costa Rica, Panama, Venezuela, Honduras, concluded negotiations with Guatemala and
-
Utilization and Innovation of Technology, 1 . pp. 30-35. ISSN 983-42733-2-3⏎This paper is a concept paper that will mainly discuss the on idea of
-
Gary White of Salisaw Equine Clinic, Salisaw, Okla. “I’m not sure whether it’s the confinement itself or just
-
Visit Event Page HERE!<|endoftext|>NORD is very grateful to M.Cristina Digilio, MD, Medical Genetics, Bambino Gesu Hospital, Rome, Italy, for assistance in the preparation of this report.⏎Syn
-
at Macquarie University, Sydney, Australia.⏎Browse the Lab Book to learn about the history, methods⏎and applications
-
km (11,100 miles) from the blast. These stations monitor seismic waves, infrasound (low frequency soundwaves), hydroacoustic, and radionuclide emissions as part of the Comprehensive Nuclear-Test-Ban
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 times that I have been thinking about lately. Let $x$ be real numbers such that $p, q, r \in \mathbb{R} $ are the product of the numbers are $p, q, r $
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
cut to the length of the object you are working on.⏎put as many folds into the ruler as you can until it gets confusing: fourths/eighths/sixteenths, thirds/sixths, square root of 3 or
-
that many jumping spiders, crab spiders, wolf spiders and other garden spiders are not indiscriminate diners, but in fact
-
the whole comes before the individual’ and the direction of .all members of society by a single will supposed to represent the ‘whole’.<|endoftext|>Ears, Nose, and Throat Surgery⏎One of the most critical areas on the human body is
-
anode bar, and spacing distances between cathode and anode of 2, 3, and 4 in. (51, 76, and 102 mm). The specimens were maintained at a constant temperature and
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
- L5:419 excites
- L5:159 weakly suppresses
- L5:531 weakly suppresses
- L6:533 weakly suppresses
- L4:182 weakly suppresses
- L4:321 weakly excites
- L4:273 weakly excites
- L4:502 weakly excites
- L2:298 weakly excites
- L4:436 weakly excites
- L4:276 weakly excites
- L5:216 weakly suppresses
- L6:136 weakly suppresses
- L5:2847 weakly excites
- L5:1435 weakly excites
- L6:2947 weakly suppresses
- L6:906 weakly excites
- L3:621 weakly excites
- L5:743 weakly suppresses
- L5:1510 weakly excites
- L5:1388 weakly suppresses
- L1:812 weakly suppresses
- L3:1222 weakly excites
- L1:734 weakly excites
Feeds later layers
- L10:554 suppresses
- L8:150 suppresses
- L10:1901 suppresses
- L9:622 suppresses
- L13:932 suppresses second digit of a year range
- L8:719 suppresses
- L14:2459 suppresses
- L12:1205 suppresses
- L13:1138 suppresses
- L13:697 suppresses
- L12:1700 suppresses
- L10:1511 suppresses
- L11:2551 suppresses
- L17:1740 suppresses
- L10:1912 weakly suppresses
- L8:336 weakly suppresses
- L14:1335 weakly suppresses
- L12:1813 weakly excites
- L9:2353 weakly suppresses
- L10:2968 weakly suppresses
- L14:2647 weakly suppresses 'and' or 'or' in split hyphen compounds
- L11:1024 weakly suppresses
- L13:2476 weakly suppresses
- L11:1960 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
- amodel
- aturally
- qp
- 皋
- \db
- cular
- bersome
- aml
- AMB
- ucing
Suppresses
- and
- Dis
- 散
- Pre
- Dis
- 닥
- 和支持
- and
- 장을
- 급