unclear: weak, word-piece and 'Note' labels
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 responds weakly to pieces of longer words, like the middle of "LAPACK" or "Zooxanthellate". It is strongly pushed down by the word "Note" at the start of a side remark, but the overall pattern is hard to pin down.
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.
-
scientific and statistic computing. A fundamental component in both NumPy/SciPy and R are the linear algebra functions, which rely on the standard BLAS and LAPACK routines. Intel® Math Kernel Library (Intel® MKL) provides a high
-
, and extends over approximately 160 km2 of the sea floor. Zooxanthellate corals are primarily Agaricia lamarcki, A. fragil
-
9; and Pueblo, 103,411.<|endoftext|>|Met proto-oncogene (hepatocyte growth factor receptor)|⏎Crystallographic structure of MET. PDB rendering based on 1r0p.⏎
-
geography teacher.⏎Is a crush love?⏎Crush is defined as a brief but intense infatuation for someone, especially someone inappropriate or unattainable. … Unlike crushes and states of infatuation, love truly sees and accepts their
-
the lesion. Collectively, DNA restoration pathways keep up with the fidelity from the human being genome. Open up in another window Shape 1 Brief Overview of DNA Damage Repair: Three common types of DNA damage are
-
is critical that they are appropriately labeled and combined into laboratory kits that are complete, well-organised, and properly targeted for temperature-controlled shipping and storage.⏎First, the purpose of the kit or kits must be firmly defined. This is a critical
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: The first thing to note is that the first two examples (The first example is the same as the first example in the second paragraph) are two different examples, but the first example in the second paragraph is the same as the first example in
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
osins transduce the chemical energy of ATP hydrolysis into mechanical motion will be very important in clinical settings (note, there are now phase II clinica trials of a small molecule activator of
-
to read about a concept to understand it.⏎- Kinesthetic leaners, who like to use movement or manipulate objects in a "hand on" approach.⏎[NOTE: The word kinesthesia means "muscle sense," which is your perception of
-
the heat and steadily converting the gas back to a liquid state. A liquid receiver captures and stores the refrigerant between the condenser and the metering device.⏎Note: Liquid receivers are not typically used on refrigeration systems, which commonly rely on
-
(CWaterlooEmitter - groundwater) / membrane wall thickness⏎where D is the diffusion coefficient for the membrane material and is expressed in L2 / t.⏎Note: In gas release applications, mass is released from the tubing on a molecular basis
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
- L2:2525 excites
- L4:2615 suppresses
- L4:19 weakly suppresses
- L2:1346 weakly suppresses
- L4:44 weakly excites
- L2:472 weakly suppresses
- L4:79 weakly excites
- L4:1876 weakly suppresses
- L3:774 weakly suppresses
- L1:2134 weakly excites
- L1:524 weakly suppresses
- L1:170 weakly excites
- L4:444 weakly suppresses
- L3:734 weakly excites
- L3:270 weakly suppresses
- L4:1052 weakly suppresses
- L4:23 weakly excites
- L3:1959 weakly suppresses
- L2:2665 weakly excites
- L4:486 weakly excites
- L0:383 weakly suppresses
- L3:913 weakly suppresses
- L3:2393 weakly suppresses
- L1:1496 weakly excites
Feeds later layers
- L6:2388 suppresses
- L17:1383 suppresses
- L21:2405 suppresses
- L15:2308 weakly suppresses
- L19:2394 weakly suppresses
- L20:733 weakly excites
- L17:2433 weakly suppresses
- L13:2556 weakly suppresses
- L22:2596 weakly excites
- L12:2442 weakly excites
- L20:957 weakly suppresses
- L7:1017 weakly excites
- L15:2265 weakly suppresses
- L15:1540 weakly suppresses
- L10:251 weakly suppresses
- L19:2831 weakly suppresses
- L16:2294 weakly suppresses
- L12:1110 weakly excites
- L12:2000 weakly excites
- L27:1034 weakly excites
- L8:1994 weakly suppresses
- L24:2782 weakly excites
- L17:2292 weakly suppresses
- L17:797 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
- ewolf
- reelection
- bows
- ptime
- GetEnumerator
- -Day
- cheapest
- icides
- طريق
- communism
Suppresses
- 注意
- Note
- note
- notes
- 要注意
- Note
- note
- (note
- >Note
- noting