capital 'I' starting an acronym
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. 22 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 a capital 'I' that begins a word or acronym in the middle of a sentence, such as 'ICREA' or 'IUCN' in the text. It is strongly negative on the start 'Di' of words like 'Diabetic' and 'Diagnostics'. Turning it up or down did not produce a clear theme.
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 study, conducted on Drosophila melanogaster by scientists at the Institute for Research in Biomedicine (IRB Barcelona) and led by ICREA researcher Marco Milán, has reproduced each of the steps known
-
(1986); the APA’s Award for Outstanding Lifetime Contributions to Psychology (1993)⏎- ACM fellow (1994)⏎- IJCAI Award for Research Excellence (1995)⏎- Katherine,
-
, where lung injury is caused by severe immune reactions, that could be a beneficial combination, said Sakshi Dua, a pulmonologist and professor at the Ica
-
a Russian MiG-15 and went in pursuit.⏎Captain Semyon Alekseyevich Fedorets, commanding the 913th IAP (Istrebitel’nyy Aviatsionnyy Pol
-
coal to back up their intermittent renewable technologies and to ensure that they have adequate power to satisfy electricity demand.”⏎This claim is based on a previous paper published by the IER itself (PDF). But as we demonstrate in our study German Coal Conund
-
the species.⏎The Great Slaty Woodpecker is a vulnerable species on the IUCN Red List, due to habitat loss. Without proper conservation measures, its
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: If we can, perhaps, open up the open door, and perhaps, perhaps, open up the open door and perhaps open open... (replaying every one of us as themselves) and perhaps open open every one of us as themselves,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, weight gain, and frothy urine.⏎What are the causes of Kidney Failure?⏎- Kidney stones⏎- Polycystic kidney disease⏎- Diabetic Nephropathy⏎- Renal tubular acidosis⏎What are
-
or internal metal pumps, these patients cannot get an MRI scan at all, as the magnet may dangerously shift these objects in the patient's body.Learn more about Diagnostics & Imaging<|endoftext|>Sudoku Puzzle #11⏎Education
-
, light sensitivity, or generalized blur from a young age. Their conditions are often progressive.⏎- Eye trauma⏎- Certain systemic diseases, like diabetes or hypertension⏎Diagnosing retinal disease:⏎Your eye doctor can check for chronic conditions like mac
-
common site of endometriosis than the small bowel. Some symptoms are due to adhesions constricting, twisting, or pulling on the bowel.⏎Diagnosis relies a great deal on the woman's description of her symptoms and menstrual history
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
- L19:1241 strongly excites
- L17:367 suppresses
- L17:366 suppresses
- L20:1964 suppresses
- L2:1463 suppresses
- L1:45 excites
- L16:118 suppresses
- L0:47 suppresses
- L16:503 suppresses
- L21:1492 excites
- L20:342 suppresses
- L0:425 excites
- L19:193 suppresses
- L2:7 excites
- L20:374 suppresses
- L10:1641 suppresses
- L14:392 weakly suppresses
- L16:711 weakly excites
- L4:221 weakly suppresses
- L6:483 weakly suppresses
- L19:311 weakly excites
- L18:1442 weakly excites unclear: word fragments and digits inside numbers
- L19:316 weakly excites
- L20:2877 weakly excites
Feeds later layers
- L27:2113 strongly suppresses
- L24:932 weakly suppresses
- L26:729 weakly excites
- L27:1716 weakly suppresses
- L27:2021 weakly suppresses
- L27:183 weakly excites
- L23:678 weakly suppresses
- L23:2087 weakly suppresses
- L27:1042 weakly excites
- L27:657 weakly excites
- L25:754 weakly suppresses
- L25:2307 weakly excites
- L26:1028 weakly suppresses
- L23:1903 weakly excites
- L25:2011 weakly suppresses linking words like 'further' and 'worse'
- L26:1634 weakly excites
- L23:2823 weakly suppresses
- L25:2906 weakly excites
- L23:1029 weakly suppresses
- L26:3042 weakly suppresses
- L27:1856 weakly suppresses
- L24:2219 weakly excites
- L25:2621 weakly suppresses
- L27:1814 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
- lay
- ging
- chin
- vy
- van
- alogy
- ISIBLE
- gen
- 真心
- a
Suppresses
- angles
- empo
- enen
- oren
- ennial
- ore
- ene
- enal
- ềm
- orest