antibiotic and drug-name context
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. 24 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 is active in medical text about antibiotics and bacteria, such as 'Ampicillin is a penicillin antibiotic'. It works against the word 'form', and it goes quiet or negative right before 'isoform' or 'form'. Turning it up made the model write unrelated sentences and turning it down made it repeat 'form'.
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.
-
extended-spectrum beta-lactamase with additional resistance to carbenicillin, amoxicillin, and increasingly to ceftazidime. The bacteria remain
-
in northern Puerto Rico. Particular attention is given to chlorinated volatile organic compounds and phthalates, although biomarkers of phenols, metals, and parabens exposure are also being explored as precursors of preterm birth. Identification of
-
floxacin is an antibiotic that is part of the fluoroquinolone family. It effectively kills bacteria, while dexameth
-
it is growing. This weakens the cell wall and it ruptures resulting in the death of the bacteria.⏎If you miss a dose of Ampicillin and are taking it regularly, take it as soon as possible. If it
-
few days.⏎Ask your health care provider any questions you may have about how to use Ampicillin.⏎Drug Class and Mechanism⏎Ampicillin is a penicillin antibiotic. It works by killing sensitive bacteria by interfering with formation of
-
affect people with underlying diseases, such as alcoholism, diabetes and chronic lung disease.⏎Treatment and resistance⏎Treatment for Klebsiella pneumonia is by antibiotics such as aminoglycosides and cephalosporins, the choice depending
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: form form form for form form for form form for form form for form form for form form for form form for form form for form form for form form for form form for form form for form form for form form for form form for form form
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
no overlap between the two diseases, the researchers noted.⏎One chaperone that protected against ALS aggregates — DNAJB14 — exists in two versions, called isoforms. The isoforms are similar, but one is shorter and lacks some important
-
Adrian Krainer. (click to enlarge)⏎One difference between metabolism in cancer and normal cells is the switch in cancer to the production of a different version, or isoform, of a protein produced from the pyruvate kinase-M (PK
-
Old French practiser or medieval Latin practizare, alteration of practicare ‘perform, perform’, from practica ‘apply’, from Greek praktikē, feminine (
-
While there are less severe and later-onset versions of citrullinemia type I, the mutations for which Counsyl screens are associated with the more severe form that affects infants shortly after birth. It is also known as “classic”
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
- L22:331 suppresses
- L19:2275 excites
- L8:1575 weakly excites
- L9:228 weakly suppresses
- L19:1251 weakly excites
- L6:2734 weakly excites
- L19:2469 weakly suppresses
- L20:1497 weakly excites
- L3:2101 weakly suppresses
- L8:611 weakly excites
- L10:2883 weakly suppresses
- L23:706 weakly excites
- L3:37 weakly suppresses
- L3:1029 weakly suppresses
- L11:2297 weakly excites
- L7:1778 weakly excites
- L15:3057 weakly excites
- L10:2954 weakly excites
- L6:2383 weakly suppresses
- L1:3039 weakly excites
- L6:1095 weakly excites
- L22:172 weakly excites
- L0:1422 weakly suppresses
- L0:2043 weakly excites
Feeds later layers
- L27:966 strongly suppresses
- L26:1710 suppresses
- L25:1400 suppresses
- L26:1407 suppresses
- L27:2151 weakly suppresses
- L25:2740 weakly excites
- L26:435 weakly excites
- L25:618 weakly suppresses
- L25:2710 weakly excites
- L25:1388 weakly suppresses
- L25:928 weakly suppresses
- L25:1170 weakly excites
- L25:62 weakly excites
- L26:349 weakly suppresses
- L26:2446 weakly excites
- L27:1535 weakly suppresses
- L26:1154 weakly excites
- L26:1254 weakly suppresses
- L26:223 weakly excites
- L26:2023 weakly excites
- L25:504 weakly suppresses
- L25:1191 weakly excites
- L25:1037 weakly suppresses
- L27:1037 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
- abay
- vements
- redits
- 委宣传
- pery
- icken
- ثقة
- aurants
- ouch
- lâ
Suppresses
- form
- forms
- form
- Forms
- Form
- Form
- forms
- _form
- →form
- _forms