'stem' in stem cell 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. 9 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 'stem' when it appears in 'stem cells', and on a few other biology words such as 'retrotransposons'. Turning it up made the model write about problems and solutions in general, which is not clearly about biology.
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.
-
will alter their functions accordingly. In embryonic development, stem cells differentiate into organs by means of the actions of morphogens. And even cancer
-
potential to grow into any cell or tissue in the body, and so can be used to replace cells that are damaged or diseased. Researchers believe the use of stem cells has the potential for curing Alzheimer's disease, Parkinson's disease, diabetes,
-
Edinburgh, showed that the DNA in some brain cells is different from the DNA in other cells due to a genetic phenomenon known as retrotransposons.⏎Retrotransposons are small bits of genetic information that can move around and multiply themselves in
-
new tissues in an embryo or regenerate damaged tissues in an adult. However, cancer stem cells are thought to hijack the ability of normal stem cells to express MET, and thus become the cause of cancer persistence and spread to
-
of U.S. scientists has identified phosphorylated signalling proteins in human embryonic stem cells (hESCs) studying which may help understand the mechanisms that determine whether stem cells divide or differentiate
-
at Children's Hospital Boston and the department of genetics at Harvard Medical School (HMS). "It also shows, from one lung cancer to another, the cancer stem cells are not the same."⏎Cancer stem cells are a subset of cancer cells
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 number 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 2
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
0.432 g of Linolenic acid.⏎Nutritional facts Per 100 Grams⏎Helps in improving vision⏎Beta-carotene is very much important for improving vision. Many antioxidants can be found in palm
-
= talk_to_server(parse_xml_document=True)If you read it out loud, the code "reads"
-
atory and human feeding studies suggest that if consumed at usual rates, GR can supply as much as 42% of the EAR for Vitamin A. Beta-carotene is only converted into Vitamin A by the human body as needed<|endoftext|>Day
-
the current scenario than Roman numerals, however Roman numerals are still useful. Arabic numerals share their origin from the numbering system of
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
- L8:2684 suppresses
- L8:2963 excites
- L8:600 excites
- L8:555 weakly suppresses
- L4:1158 weakly suppresses
- L6:16 weakly excites
- L5:369 weakly suppresses
- L4:2114 weakly suppresses
- L7:2892 weakly suppresses
- L8:2312 weakly suppresses
- L4:1476 weakly suppresses
- L7:2271 weakly excites
- L5:1214 weakly suppresses
- L7:1675 weakly excites
- L6:17 weakly suppresses
- L7:2963 weakly excites
- L6:1628 weakly excites
- L5:1438 weakly excites
- L0:1464 weakly suppresses
- L7:2648 weakly suppresses
- L7:2698 weakly excites
- L7:39 weakly excites
- L6:2973 weakly suppresses
- L7:1661 weakly suppresses
Feeds later layers
- L10:2678 excites
- L17:2487 suppresses
- L10:2052 suppresses
- L10:2186 suppresses
- L17:1758 suppresses
- L11:201 suppresses
- L10:1084 excites
- L11:1200 weakly excites words that signal persistence (still, retains, continue to)
- L10:1357 weakly excites
- L18:1848 weakly excites
- L20:1717 weakly excites
- L15:1562 weakly excites
- L10:277 weakly suppresses
- L17:1970 weakly excites
- L22:2861 weakly excites
- L11:1442 weakly suppresses
- L12:3058 weakly suppresses
- L12:759 weakly suppresses
- L11:1928 weakly excites
- L17:2383 weakly excites
- L18:1450 weakly suppresses
- L12:703 weakly excites
- L22:2232 weakly excites
- L17:1957 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
- icol
- colspan
- cca
- 客户提供
- ��
- ]=[
- ander
- 抵
- ;c
- Qi
Suppresses
- .parseFloat
- estring
- .parseInt
- -parse
- later
- ată
- emouth
- parse
- eled
- Later