'his team' and 'colleagues' in research news
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. 5 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 the word for the group of researchers in science news, like 'his team', 'her colleagues' or 'scientists' after 'with' or 'When'. It seems to anticipate a group of scientists being named. Turning it up gave unrelated text, so the steering results don't show this.
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 work on the structure of the virus shells and identification of mutations to improve their stability was carried out by Professor David Stuart and his team at Oxford University using Diamond Light Source, the UK's national<|endoftext|>Coronavirus
-
which is a hallmark of cancer and has a well-established role in aggressive B cell malignancies.⏎To find out more, Hans-Guido Wendel and colleagues from Memorial Sloan-Kettering Cancer Center in New York analyzed genomic data from two
-
neuron bundles in the spinal cord. It is not quite clear how these central pattern generators know how quickly the legs are to be moved.⏎An international team working with scientists from Harvard University and the Max Planck Institute of Neurobiology in Martinsried has
-
such as algae, off rocky surfaces. The teeth are built of protein frames reinforced by closely packed nanofibers of a mineral called goethite.⏎When scientists attempted to pull this light structure apart in the lab, it displayed a very high
-
impact on ancient societies in warmer climates, such as those in the Eastern Mediterranean, thanks to the degradation of DNA in the higher temperatures.⏎So Neumann and his team went digging through bones recovered from a site on Crete known for its remarkably cool
-
cell biology with genetic typing to identify what drives the cancerous behavior of each patient's tumor and to develop new therapeutic targets.⏎In their study, Kim and her colleagues looked at mouse models of the three most commonly mutated genes in human lung cancer --
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 to be sure of my own life, before I begin to sing my song 'In the Weeping'... I think, I have, I will sing this song... and so I repeat, I will sing this song... and
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
- Filed Under⏎J ennifer Skaja has seen the signs of small miracles, as children with disabilities gain their voice through American Sign Language. Individuals with hearing impairment have⏎long used sign language to communicate, but studies
-
ron schedule file. The advantage of this approach is the browser fires off a seperate HTTP request, corresponding to a seperate Apache process, so any code executed by pseudocron is run “out of band” from the main PHP script – visitors to
-
February 2011, the UN Special Representative to Kosovo called for an independent investigation into Kosovan organ trafficking allegations to be spearheaded by the UN Security Council.⏎Noting the changing forms of slavery, including the rise in global sex trafficking
-
of funding,” Turner said.<|endoftext|>This article touches on cultural, linguistic, and technical issues that arise in video recording stories of Deaf individuals who use American Sign Language (ASL). Whether you are Deaf or hearing, documenting ASL interviews
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
- L3:1133 suppresses
- L3:731 weakly suppresses
- L2:2210 weakly suppresses
- L4:2479 weakly suppresses
- L4:2764 weakly excites
- L3:2589 weakly excites
- L2:438 weakly suppresses
- L2:1996 weakly suppresses
- L0:1639 weakly suppresses
- L4:1893 weakly excites
- L4:2536 weakly suppresses
- L3:1010 weakly suppresses "to" in "click here to"
- L3:1542 weakly excites
- L0:2875 weakly suppresses
- L3:2591 weakly suppresses
- L4:1916 weakly suppresses
- L2:2586 weakly excites
- L4:2134 weakly excites
- L1:2885 weakly suppresses
- L0:1249 weakly suppresses
- L0:1105 weakly suppresses
- L4:892 weakly excites unclear: weak math-topic words
- L4:29 weakly suppresses
- L4:2822 weakly suppresses
Feeds later layers
- L11:1423 excites
- L6:337 excites
- L7:1111 suppresses
- L7:267 weakly excites
- L19:1626 weakly excites
- L7:34 weakly suppresses
- L17:1677 weakly excites
- L7:1034 weakly excites
- L12:151 weakly excites
- L12:2796 weakly excites
- L10:2608 weakly excites
- L9:1754 weakly excites
- L22:1460 weakly excites
- L7:590 weakly suppresses
- L19:2865 weakly excites
- L9:718 weakly excites
- L20:592 weakly excites
- L9:1494 weakly suppresses
- L8:2198 weakly suppresses
- L16:1625 weakly excites
- L9:481 weakly suppresses
- L13:2544 weakly excites
- L13:1621 weakly suppresses
- L24:551 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
- [at
- urities
- олод
- ethylene
- atform
- ")] ⏎
- iculty
- uetype
- 攻击力
- 利用您的
Suppresses
- reg
- REG
- Reg
- rut
- regular
- 准则
- regulated
- faf
- nu
- .reg