the word 'if'
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 fairly consistent: several of them 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 strongly on the word 'if', especially when it starts a sentence or stands alone as a word. When it was turned down, the model got stuck repeating 'if if if', which fits the idea that it pushes against that word.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
Key words: tuna / Fish Aggregating Devices / behaviour / horizontal movements / ultrasonic telemetry / French Polynesia⏎© Elsevier, Inra, Ifremer, Cemagref, Ird, Cnrs,
-
James McIntyre⏎Read this poem in other languages⏎This poem has not been translated into any other language yet.⏎Still I Rise⏎The Road Not Taken⏎If You Forget Me⏎Edgar Allan Poe⏎Stopping By Woods On A Snowy
-
, New York. It helped to launch the U.S. campaign for women’s suffrage, or the right to vote.<|endoftext|>Have you ever wondered?⏎- If animals such as giraffes and
-
‘raw’ in ‘straw’ or ‘if’ in ‘life’, as they are bound to appear double in the playing field
-
natural by-product of gas<|endoftext|>DataTableCollection.Item Property (String)⏎Gets the DataTable object with the specified name.⏎Assembly: System.Data (in System.Data.dll)⏎If a DataTable name exists that matches the case of the search string, it is returned
-
of Logical Knowledge. M.D.S. Braine, The "Natural Logic" Approach to Reasoning. E.K. Scholnick, The Three Faces of If. F.B. Murray, The
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 if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if if
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
auditing, billing, and logging. In addition, the architecture also offers enterprises the flexibility of having agile business processes, better addresses the regulatory requirements like Sarbanes Oxley (SOX), and changes individual services without affecting other services.⏎The reality
-
part of the body in less than 72 hours. MRSA is strong - resistant to nearly all antibiotics, including penicillin, amoxicillin, oxacillin and methicillin. It causes more deaths in the U.S.
-
muscles, sensations, behaviour, emotions, consciousness, or a combination of these. The seizures can be prevented in most people by suitable antiepileptic medication. Oxcarbazepine works by stabilising the electrical activity of your brain, which
-
it is added to other white powders to increase dealers' margins and users' highs. While deaths due strictly to the abuse of prescription opioids, such as OxyContin, peaked around 2017, fentanyl-involved deaths have
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
- L19:2348 excites
- L17:1282 excites
- L0:1470 excites
- L6:2134 excites
- L18:1617 excites
- L9:306 suppresses
- L20:466 weakly suppresses
- L17:619 weakly excites
- L19:1580 weakly excites
- L15:851 weakly excites
- L9:75 weakly excites 'what' or 'how' after see or understand
- L20:342 weakly suppresses
- L13:2966 weakly suppresses
- L1:1850 weakly suppresses percent-encoded url characters
- L0:921 weakly excites
- L15:740 weakly excites
- L20:2699 weakly excites
- L4:265 weakly suppresses
- L3:2440 weakly suppresses
- L11:502 weakly excites
- L8:132 weakly suppresses
- L18:1389 weakly excites
- L14:1324 weakly suppresses
- L16:1564 weakly excites
Feeds later layers
- L23:298 excites
- L26:509 suppresses
- L23:449 suppresses
- L27:511 weakly suppresses
- L25:768 weakly suppresses
- L26:1599 weakly suppresses
- L27:2280 weakly excites
- L24:1474 weakly suppresses
- L26:217 weakly suppresses
- L25:234 weakly excites
- L26:715 weakly excites
- L25:2302 weakly suppresses
- L24:123 weakly suppresses
- L23:1815 weakly suppresses
- L27:717 weakly excites
- L23:2750 weakly suppresses
- L24:1581 weakly suppresses
- L23:999 weakly suppresses "be" after modal verbs
- L26:466 weakly excites
- L27:2219 weakly excites
- L27:2209 weakly suppresses
- L24:2631 weakly suppresses
- L24:2423 weakly suppresses
- L23:1068 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
- EVER
- possible
- DebugEnabled
- {[
- orgot
- you
- POSS
- ighting
- 您
- 吒
Suppresses
- 如果
- 如果有
- 但如果
- ificantly
- 如果不是
- 如果我们
- 如果
- if
- ificate
- ivement