'yet' and 'haven't' after have
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. 4 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 words like 'yet', 'haven't' and 'have now' that show something has or has not happened so far. The examples where it is strongly negative are mostly the name Ramses, but that is a weaker pattern. Turning it up did not give a clear theme in the replies.
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.
-
of existing ties.⏎In addition, volunteering can be a risk-free method to try out a new profession. If you're thinking about a career in nursing but have yet to gain much experience, volunteering
-
. We can multiply by 5 and divide by 5.⏎- 1/5 times 5 is just going to be 1. So we haven't changed the value of the expression.⏎- But when we do it this way,
-
commemorate the Jewish holiday, Yom Ha’ Shoah, will begin at 5 p.m. in the Hatfield
-
school-business partnership.⏎- Design & Thinking Official Trailer (02:28)⏎Design & Thinking is a documentary released this year about design thinking. I haven't seen it yet, but I'm loving the approach -- the cacophony of
-
shield against certain kinds of flu virus.⏎Some individuals suggest drinking Echinacea tea to shorten the length of cold and respiratory illness symptoms. However, this result has yet to be tried by research.⏎As we know garlic
-
Yom Ha' Shoah is not just a Jewish event — it touches everyone,” she says.⏎“Yom Ha' Shoah is a wake-up call that genocide is happening now and we have
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 power of a king in the capital: the god of a god of the goddess of a goddess. The king is in the city of a. The god of a god of a. The goddess of a. The queen of a.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
from being constructed to continue the cult of the deceased Pharaoh and to honor Amon, was also built in reverent memory of Seti's father, Ramses I, who died before constructing a temple of his own. It was not completed
-
1190–87 bce), founder of the 20th dynasty (1190–1075 bce), Ramses found Egypt upon his accession only recently recovered from the unsettled political conditions that had
-
, instead of Apap, the serpent of darkness, the great antagonist of the beneficent Osiris. The hatred of the Egyptians for the very name of Set was carried so far that it was chiseled out of the monuments; the
-
and Mount Carmel. The attempted invasion ended Egyptian pretensions to a Syro-Palestinian hegemony.⏎Two more years of peace ensued, but in Ramses’ 11th year a new coalition of Libyan tribes infiltrated the western
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
- L3:464 excites
- L3:1933 excites
- L1:2444 weakly suppresses
- L2:70 weakly excites
- L3:421 weakly suppresses
- L3:2369 weakly suppresses
- L3:1496 weakly excites period in abbreviations and initials
- L3:50 weakly excites
- L2:75 weakly suppresses
- L3:512 weakly suppresses
- L1:743 weakly suppresses
- L3:691 weakly excites
- L3:2351 weakly suppresses
- L3:953 weakly suppresses
- L3:769 weakly suppresses
- L0:1416 weakly suppresses
- L1:180 weakly suppresses
- L3:1260 weakly suppresses
- L3:2440 weakly suppresses
- L3:2501 weakly excites
- L3:1633 weakly excites
- L3:2473 weakly suppresses
- L2:1321 weakly suppresses
- L2:3042 weakly suppresses
Feeds later layers
- L5:660 excites
- L10:817 suppresses
- L20:1632 suppresses
- L11:1926 suppresses
- L5:2738 suppresses
- L5:214 weakly suppresses
- L5:2623 weakly suppresses
- L22:2657 weakly suppresses
- L6:2001 weakly suppresses
- L9:2449 weakly excites
- L17:1931 weakly suppresses
- L13:1295 weakly suppresses
- L18:911 weakly excites comma after 'and' or 'and then' clause
- L5:17 weakly suppresses
- L10:1150 weakly suppresses
- L9:629 weakly excites
- L16:2656 weakly excites
- L7:1144 weakly excites
- L14:2875 weakly suppresses
- L19:2948 weakly excites
- L12:2697 weakly excites
- L21:144 weakly excites
- L21:1281 weakly suppresses
- L11:1714 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
- ivant
- <Class
- categor
- المادة
- MX
- here
- here
- 惯
- \TestCase
- (class
Suppresses
- 谁能
- 一族
- leneck
- 所能
- .cloudflare
- щения
- ucken
- 报仇
- çıktı
- ились