'for' in research institute names
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. 3 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 word 'for' inside the names of institutes and organisations, such as 'Institute for Research' or 'Institute for Animal Health'. Turning it up did not produce anything obviously related, so the steering test tells us little about it.
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.
-
variety of reasons, including the study of human genetic diseases,'' said Dr. Gregory M. Acland, a research scientist at the James A. Baker Institute for Animal Health at
-
the plan doesn’t answer is how the city will cope with the climate change that has already been set in motion.⏎Enter a study prepared by Oakland’s Pacific Institute for the California Energy Commission, published in July but not widely circulated until this month.
-
Other agents have also been moved to the highest risk category following the report.⏎The findings that led to the decision by the World Health Organization (WHO) International Agency for Research on Cancer (IARC) to reclassify the cancer risk of sunbed
-
and trusted source of scientific evidence of what works in education. The WWC is administered by the Department, through a contract to a joint venture of the American Institutes for Research and the Campbell Collaboration.⏎To provide information needed by decision-makers, the W
-
four times more likely than older drivers to crash for every mile driven.⏎The graduated licensing programs are meant to prevent this.⏎James Fell and colleagues at the Pacific Institute for Research and Evaluation in Maryland examined data on fatal crashes from 1990
-
Linzhao Cheng, Ph.D., an associate professor of gynecology and obstetrics and co-director of the stem cell program of the Johns Hopkins Institute for Cell Engineering. “This is one area of stem cell biology where human and mouse
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: if I have a problem of a book, a book, a book, and I have a problem, a problem, a problem, and a problem. I have a problem of a book, a book, a book, and I have
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, It is too quiet on the streets below.⏎2. I wish we were not stuck in the house all the time, said Winnie.⏎3. Wilbur did not like his special collar.⏎4. The scientist on the TV said you
-
"⏎- Brix Orolfo, age 12yrs.⏎- © Brix Orolfo⏎
-
Ocean. Ukrainian Cossacks formed the Zaporozhia Cossack State in 1649. They are considered
-
office for women at the State Department, to Amina Benkhadra, Morocco’s Minister of Energy, Mines, Water and Environment, to Christiana Figueres, executive secretary of the U.N. framework on Climate Change, women are
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
- L0:1543 excites
- L2:1622 weakly excites
- L1:1341 weakly suppresses
- L1:2581 weakly excites
- L1:1258 weakly excites
- L0:1474 weakly excites
- L0:340 weakly suppresses
- L1:1179 weakly excites 'in' after 'result' (results in)
- L0:1939 weakly suppresses
- L1:1710 weakly excites
- L2:2422 weakly suppresses
- L2:2351 weakly suppresses
- L1:813 weakly suppresses
- L0:2044 weakly excites
- L1:529 weakly excites
- L2:1539 weakly suppresses
- L1:1793 weakly suppresses 'he' in pheasant, 'mouth' in Dartmouth
- L0:644 weakly suppresses
- L2:500 weakly suppresses
- L1:2740 weakly suppresses word fragments inside non-English words
- L2:2558 weakly suppresses
- L0:694 weakly suppresses 'seen' after 'never' or 'rarely'
- L2:2656 weakly excites
- L2:1321 weakly suppresses
Feeds later layers
- L19:1221 strongly excites
- L5:1090 excites
- L17:1223 excites
- L16:2336 excites
- L16:524 excites
- L9:2587 excites
- L10:1029 excites
- L7:1172 excites 'for' after past participles like selected
- L4:1035 excites 'for' after verbs like 'fall' or 'click'
- L4:1653 excites
- L18:1028 excites
- L16:463 excites 'for' introducing a purpose or target
- L17:2949 weakly excites 'for' introducing an example or reason
- L16:2690 weakly suppresses
- L23:2014 weakly excites
- L19:1237 weakly excites
- L23:2386 weakly excites
- L18:1818 weakly suppresses quantifier 'all' mid-phrase
- L9:1890 weakly excites
- L21:474 weakly suppresses
- L7:2026 weakly excites
- L18:2783 weakly excites
- L6:1880 weakly suppresses "by" introducing a deadline or means
- L15:2925 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
- geries
- lops
- purposes
- bidden
- 目的
- amen
- yı
- tran
- andes
- pastoral
Suppresses
- egot
- ENSIONS
- eña
- ᴇ
- �
- ess
- 意义上
- e
- ENSION
- 以习近平