word pieces inside medical and science terms
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 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 the middle pieces of longer science and medical words, such as 'amino' in 'amino acid', 'planing' and 'manometry'. It also reacts to the letter 'n' in sample-size notation like 'n=34'. The steering results were mixed and did not show a clear theme.
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.
-
the form of polymeric chains in which successive ubiquitin molecules are connected through specific isopeptide bonds. Once attached to a substrate, the 76-amino acid ubiquitin is subjected to further modifications, creating a multitude of distinct signals with
-
’s teeth and gums during an exam to choose the best gingivitis treatment option. Most likely, he will recommend a deep cleaning and/or scaling and root planing for more severe cases<|endoftext|>Considering the fact that half of you reading this right now
-
accumulation of evidence supporting a major role of these proteins in the regulation of various biological and pathological phenomena.⏎Ubiquitin (2) is a 76-amino acid protein, which is highly conserved across all eukaryotes. Remark
-
On physical exam the leg has no erythema, but does have mild tenderness along the distal tibial shaft. Plain radiographs are negative. What is the most appropriate management?⏎vitamin D and calcium levels
-
tone and strength.⏎The strength of the pelvic floor is not always easy to assess through physical exam, but a test called anorectal manometry can easily quantify it. Another test, called EMG, can determine the nerve
-
02, 2012⏎Repairing bones with nanomaterials⏎(Nanowerk News) Around half a million people in the EU need treatment for serious
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: 1000. It is a number with 1000. The number of 1000. 1000. 100. 1000. 100.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
2016, figure 10.4, page 207)⏎The prevalence of these types of trauma in the Late Intermediate Period (n=4/8
-
meeting the CCHR criteria excluded. Ten parameters were extracted and tabulated. Results: Forty-one patients were included with three exclusions. 89% (n=34) of the patients presented with a 2-hour GCS of
-
v. To use; to exercise; to inure; to accustom by practice.⏎from the GNU version of the Collaborative International Dictionary of English⏎- n. Use; practice; exercise.⏎- transitive v. To use; to
-
s,⏎repeated measures design counterbalanced to prevent order effects will⏎be used. Small quantities of JA (two per week) are provided to persons⏎(n = 35) with NIDD for two weeks at a time alternating with two
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
- L4:2298 excites
- L4:1409 weakly suppresses
- L3:1278 weakly suppresses
- L4:2902 weakly suppresses
- L3:896 weakly excites
- L3:1675 weakly suppresses
- L2:1937 weakly suppresses
- L0:2346 weakly excites
- L2:1856 weakly excites
- L1:2234 weakly excites
- L3:3029 weakly suppresses
- L4:1410 weakly excites
- L3:1713 weakly suppresses
- L3:901 weakly suppresses
- L3:271 weakly excites
- L3:1239 weakly excites
- L4:1503 weakly suppresses
- L0:1950 weakly suppresses
- L2:2047 weakly excites
- L4:1310 weakly suppresses
- L0:2621 weakly excites
- L4:1080 weakly suppresses
- L2:2602 weakly suppresses
- L2:2331 weakly excites
Feeds later layers
- L8:1224 suppresses
- L10:2474 excites
- L6:896 weakly suppresses
- L23:1370 weakly excites
- L13:2000 weakly excites
- L7:2226 weakly excites
- L12:2077 weakly excites
- L7:1095 weakly excites
- L22:2745 weakly suppresses
- L23:407 weakly excites
- L6:3011 weakly excites 'in' after a named author or speaker
- L24:636 weakly suppresses
- L6:2880 weakly suppresses
- L20:1032 weakly excites
- L23:517 weakly excites
- L23:1877 weakly suppresses
- L19:1873 weakly suppresses
- L19:2101 weakly suppresses
- L9:1631 weakly suppresses
- L20:1558 weakly suppresses
- L7:2150 weakly suppresses
- L6:1529 weakly suppresses
- L10:1614 weakly suppresses
- L18:1089 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
- pecified
- •⏎⏎
- Illustrated
- rangle
- pic
- 藏着
- obs
- <tbody
- __(*
- scor
Suppresses
- enance
- ưỡng
- .PREFERRED
- dez
- regenerate
- ="?
- ariate
- gregate
- 一事
- cleanly