'other' after 'any', 'all' or 'or'
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. 15 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
Fires on the word 'other' when it follows 'any', 'all', 'or' or 'and', as in 'any other platform' or 'all other IBM services'. The pattern is very consistent in the examples. Turning it up made the model repeat phrases like 'a single person', which is only loosely related.
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.
-
Fraud⏎Mail fraud describes any action meant to defraud somebody that makes use of the United States Postal Service or any other interstate mail carrier. This would include any action in which somebody tries to cheat a
-
from a laboratory test. Our Environmental Health professionals will ensure testing of animals that could be infected with rabies. To stay safe, never handle a bat or any other wild animal” stated Erie County Health Commissioner, Dr. Gale R. Burstein
-
share the road with, motorcycle riders. A motorcyclist has the same rights, privileges, and responsibilities as any other motorist on the roadway.⏎“Motorcyclists will be out in force as
-
, and Scala so that you start creating your Spark projects and collaborating with other data scientists. When you sign up, you get free access to Watson Studio and all other IBM services for 30 days. Start now and take advantage of this offer
-
to finance its assets. The WACC is the minimum acceptable return that a company must earn on an existing asset base to satisfy its creditors, owners, and other providers of capital, or they will invest elsewhere.⏎The following is the WACC
-
to upload the pdf to Brightspace or Google Classroom (or any other platform you are using):⏎Have students record answers on top of the pdf⏎Yes
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 previous year, I was in a situation where I was in the of my previous relationship and it is what I thought of about the relationship. I was, and I have these things in my mind. I was what for the relationship.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
is available at http://www.usdoj.gov/crt/housing/housing_rluipa.htm. Below is a summary of the law’s key provisions, with illustrations of the types of cases that may violate the law.⏎⊕ RLUIPA
-
"focus" uses 5 letters: C F O S U.⏎No direct anagrams for focus found in this word list.⏎List shorter words within focus, sorted by length⏎All words formed from focus by changing one letter⏎Browse words starting
-
and begin to code.⏎Your program should read from the file . The file will describe a single post-cubist squarist
-
we said goodbye to four retired NASA space shuttles that found new museum-type homes.⏎Here's our list of the biggest science achievements this year, in order of significance:⏎1. Curiosity lands, performs science on Mars⏎Every time I
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
- L11:1809 excites
- L9:792 weakly suppresses
- L14:2808 weakly suppresses
- L10:720 weakly suppresses
- L6:2600 weakly excites
- L12:1840 weakly excites
- L12:1853 weakly excites
- L6:2473 weakly suppresses
- L12:3027 weakly excites 'other' in 'and other ...' lists
- L13:1105 weakly suppresses
- L13:2818 weakly excites
- L2:730 weakly suppresses
- L12:942 weakly suppresses
- L14:383 weakly suppresses
- L7:1587 weakly suppresses
- L5:1944 weakly excites
- L1:1016 weakly suppresses
- L8:2280 weakly suppresses
- L10:49 weakly excites
- L2:1931 weakly excites
- L13:167 weakly excites
- L7:2552 weakly excites
- L9:2383 weakly suppresses
- L14:2904 weakly suppresses
Feeds later layers
- L16:1984 weakly suppresses
- L18:1738 weakly suppresses
- L24:236 weakly excites
- L17:2200 weakly suppresses
- L20:1156 weakly suppresses
- L21:676 weakly excites
- L16:2087 weakly suppresses
- L18:2426 weakly suppresses
- L18:679 weakly suppresses
- L16:327 weakly excites
- L20:2141 weakly excites
- L20:1271 weakly excites
- L17:2670 weakly excites
- L17:2331 weakly suppresses
- L16:2620 weakly suppresses
- L25:1152 weakly suppresses
- L24:521 weakly suppresses
- L16:1969 weakly suppresses
- L24:614 weakly suppresses
- L16:2818 weakly suppresses
- L19:2560 weakly excites
- L16:2040 weakly excites
- L20:2275 weakly excites
- L19:2989 weakly suppresses
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
- possible
- asters
- ажд
- quires
- itest
- quier
- aches
- Possible
- _other
- ills
Suppresses
- defensively
- --------⏎
- IBC
- IFIER
- dad
- 态
- ère
- Lia
- latina
- .lastIndexOf