unclear: prepositions in science prose
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. 20 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 small connecting words like 'in', 'of' and 'into' that sit between a noun and what comes after it, mostly in biology and genetics writing. Turning it up made the model's replies drift into random topics, so it doesn't show a clear effect on meaning.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
more or less susceptible to disease is one of the biggest challenges in science today. Geneticists study how different genetic profiles affect how certain genes are turned on or off in different people, which could be the cause of a number of genetic disorders.⏎L
-
require a student to be familiar with a lot<|endoftext|>A selectable marker is a reporter gene introduced into a cell along with a gene insert. What this means is, the experim
-
within the last decade have allowed us to do so much more. We can now conduct whole-exome sequencing, a process that directly scrutinizes all protein coding regions of the genome, which is approximately 1 to<|endoftext|>- About LENS⏎-
-
today.⏎The human genome contains more than 20,000 protein-coding genes, which carry the instructions for proteins essential to the structure and function of our cells, tissues and organs. Some of these genes are very similar to each
-
GMO" about this. These are breeding techniques and processes that were developed *before* the idea of genetic engineering which is the biochemical creation of specific advantageous mutations in the genome of a plant. Before the advent of GMO farmers and plant breeders
-
an odd shape in the mutant mice: Their axons and dendrites are shorter and have fewer branches than those
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 a very important variable, which is a variable that is not present on the original variable. I have a variable that is a variable that is not present on the original variable, and I have a variable that is a variable that is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
turned out to be hundreds or thousands of variants involved in common diseases, and it was unclear what each one did. Only about 5% of the shared variants were in sections of DNA forming the template for genes.⏎“We were missing too much information
-
, coronary artery disease and bipolar disorder.⏎Based on the characteristics of the variants, and their location on the genome, they could tell whether the disease-causing traits are increasing or decreasing over time.⏎"We asked the question: When we look at human
-
determine this, so let’s take a look.⏎What Actually Is Autosomal And How Does It Work?⏎Autosomal, in simple terms, is a chromosome that can be tested to find segments of
-
000 protein-coding genes, which carry the instructions for proteins essential to the structure and function of our cells, tissues and organs. Some of these genes are very similar to each other because, as the genomes of humans and other mammals evolve
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
- L10:1864 suppresses
- L17:958 suppresses
- L8:2563 excites
- L19:806 excites
- L7:2433 weakly suppresses
- L15:1981 weakly excites
- L19:1900 weakly excites
- L18:1624 weakly excites
- L9:1845 weakly excites
- L6:2532 weakly excites
- L9:2708 weakly suppresses
- L0:1900 weakly excites
- L19:2385 weakly suppresses
- L9:2066 weakly suppresses
- L19:2289 weakly excites
- L8:911 weakly excites
- L7:2660 weakly excites
- L19:1550 weakly suppresses
- L9:2732 weakly excites
- L5:2149 weakly suppresses
- L19:1926 weakly excites
- L19:599 weakly excites
- L15:737 weakly suppresses
- L17:941 weakly excites
Feeds later layers
- L25:399 suppresses
- L24:1469 suppresses
- L25:1139 suppresses
- L22:1297 suppresses
- L25:3008 weakly suppresses
- L25:1518 weakly suppresses
- L22:720 weakly suppresses
- L21:2858 weakly suppresses
- L24:416 weakly suppresses
- L27:2798 weakly suppresses
- L22:803 weakly suppresses
- L24:570 weakly suppresses
- L21:2786 weakly excites
- L24:2514 weakly suppresses
- L21:565 weakly excites
- L21:2885 weakly suppresses
- L21:401 weakly suppresses
- L21:1753 weakly excites
- L21:2137 weakly suppresses
- L21:1058 weakly suppresses
- L21:2934 weakly excites
- L22:2632 weakly suppresses
- L22:2665 weakly excites physical measurements and dimensions
- L21:2400 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
- يات
- odel
- iology
- emics
- iors
- UnitOfWork
- succ
- ще
- stances
- iples
Suppresses
- 变量
- variable
- Variable
- 변수
- variable
- variables
- Variables
- _variable
- parameter
- _Field