"this" in time and context phrases
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 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 lights up on the word "this" when it starts a phrase about the current moment or situation, like "this week", "this autumn" or "this case". Turning it up didn't produce a clear theme in the written text, and the model drifted into unrelated topics.
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 Columbia Daily Tribune the drought has been hard on the prized specimen.⏎"The drought of '80 was different," he said. "That was hotter, but this one is longer."⏎The U.S. Drought Monitor this week said more than
-
a half century ago.⏎“Just watching that moment was pretty dog-gone exciting,” he said. “It was just like, ‘wow,’ can you believe this is actually happening?”⏎It did finally happen.⏎After years of space exploration,
-
sparking new actions that will help benefit individuals with autism, and their families, in the years to come.⏎April is World Autism Awareness Month, and three years ago this week, OHSU and more than 20 other sites across the U
-
seems as if every time Israel defends itself, somebody points an accusing finger and yells “Genocide!” Raphael Lemkin, who coined that term seventy years ago this autumn, would have been appalled by such abuse of his life’s work.⏎
-
in Seattle, who heads the UNOS living donor committee.⏎"We really do need to take a look at things again in real time to say, OK, in this current climate what are our risks?"⏎More than 90,000
-
girls have a difficult time getting an education. The sons of these workers — very much like the sons of their richer counterparts — are treated like kings, or in this case pauper-kings. They get the lion’s share of whatever little food
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: I have been thinking about... I have been thinking about... I have been thinking about... I have been thinking about... I have been thinking about... I have been thinking about... I have been thinking about... I have been thinking about...
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
anam. In that the poet relates the life story of one Punithavathiar who abandoned the
-
steady progress toward the human exploration of the solar system and beyond. In that expectation, we were cruelly disappointed. Partisan politics, a collective failure of
-
joined the Loyola community in 2013 and currently serves as the Senior Associate Director in the Academic Advising and Support Center (AASC). In that role, she advises undergraduate students and works closely with the transfer student population, assisting
-
September 10, then received the name of Hurricane Humberto at the first report of the National Hurricane Center (NHC) the next day. At that time the hurricane was located about 310 mi (500 km)
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
- L2:1590 weakly excites
- L3:108 weakly suppresses
- L3:2071 weakly excites
- L2:383 weakly excites
- L1:335 weakly excites
- L2:2476 weakly excites
- L0:2836 weakly excites
- L3:1068 weakly excites "that" pointing back to a noun
- L2:2677 weakly suppresses
- L2:146 weakly excites unclear: 'lesson' and 'lique' word pieces
- L0:2228 weakly excites
- L1:886 weakly excites
- L2:619 weakly excites
- L3:301 weakly suppresses
- L1:1339 weakly suppresses
- L2:1012 weakly excites
- L3:2641 weakly excites
- L0:1889 weakly suppresses
- L2:3048 weakly excites
- L1:2429 weakly suppresses
- L3:512 weakly suppresses
- L2:977 weakly excites
- L2:2511 weakly suppresses
- L3:745 weakly excites
Feeds later layers
- L5:98 strongly excites
- L10:2178 excites 'this' after a verb or preposition
- L9:82 excites
- L6:793 excites 'this' after 'earlier' or 'in'
- L5:89 suppresses
- L10:1104 suppresses
- L6:41 weakly suppresses
- L17:2464 weakly suppresses
- L10:965 weakly suppresses the word 'that' introducing a clause
- L19:2524 weakly suppresses
- L10:864 weakly excites
- L7:2602 weakly excites
- L17:2615 weakly excites
- L19:2916 weakly suppresses
- L12:1379 weakly suppresses
- L5:2733 weakly excites
- L13:1221 weakly suppresses comma or verb after a researcher or study subject
- L6:2404 weakly suppresses
- L18:2604 weakly suppresses
- L23:2687 weakly excites
- L21:1563 weakly suppresses
- L14:228 weakly suppresses
- L6:1445 weakly suppresses
- L18:1785 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
- cat
- @Web
- ORT
- 司
- 系列
- gif
- cat
- onet
- arti
- Figure
Suppresses
- 除外
- ..."); ⏎
- 瑕
- .'); ⏎
- balances
- ...');⏎
- 确
- '] ⏎
- 玒
- .existsSync