technical word fragments, splicing and isotopes
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. 6 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 reacts to parts of longer technical words, like 'splicing' in biology, 'isotopes' in physics, or 'allotropes' in chemistry. Turned up, the writing stayed vague and drifted into numbers and circuits.
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.
-
⏎on the Hidden Markov Model of probabilities (HMMgene) also can locate donor and acceptor sites.⏎- Splice variants: Alternative splicing, resulting in different combination of exons, has been found to occur⏎in
-
carbyne, an infinite sp1 hybridized carbon chain. As a result, much less has been explored about carbyne as compared to other carbon allotropes such as fullerenes, nanotubes and graphene. Although end-c
-
disrupted, an alarm will sound. It is most effective to warn about a smoldering fire. Ionization smoke detectors contain a very small amount of americium-241 within an ionization chamber. They create an electric current between two
-
device that at both ends converts gases of deuterium and tritium (isotopes of hydrogen) into plasmoids - plasma contained by a magnetic field through
-
⏎is derived from internal sequences called "exons" which alternate with the introns in the longer, nuclear mRNA.⏎(Figure of splicing mechanisms)⏎Splicing of nuclear mRNA is catalyzed by specialized RNA-protein complexes called small nuclear
-
pressure processes which are descriptive of "ballotechnic" reactions, i.e. shock- induced reactions in heterogeneous ma- terial mixtures.⏎- ISOTOPIC SEPARATION AND ENRICHMENT⏎- “ballotechnic”
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: analog audio, 12 to 12 ohm, and this is the audio output for a speaker. You can use a 12 ohm resistor to connect to a 12 ohm resistor directly. This is a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
4. Why is data bus bi-directional?⏎5. What is the function of accumulator?⏎6. What is flag, bus?⏎7. What are tri-state devices and why they are essential in a bus oriented system?⏎8. Why are
-
with 1/4" TS and TRS connectors, XLR connectors, and other plugs and jacks. Headphone connectors and other 1/8" plugs are also TRS connectors, where the two signals share a common ground connection
-
.6µm Bipolar Complementary Metal Oxide Semiconductor, or BiCMOS, is typically implemented as dual NAND gates with coupled inputs and a BJT driving the output. BiCMOS is an adaptation of older CMOS technologies, and
-
video about signal flow,…there are three main categories…of the strength or level…of the electrical analog signal.…⏎Mic level, line level, and speaker level.…Mic level is the weakest of the three.…This is usually somewhere around…
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
- L4:2392 excites
- L4:1478 suppresses
- L4:1968 excites
- L2:1670 weakly excites
- L1:1340 weakly suppresses
- L5:46 weakly excites
- L4:2941 weakly excites
- L4:1066 weakly excites
- L4:2254 weakly suppresses
- L0:2583 weakly suppresses
- L5:1462 weakly excites
- L3:322 weakly excites
- L0:2464 weakly suppresses
- L5:707 weakly suppresses
- L1:2922 weakly excites
- L4:112 weakly suppresses
- L5:216 weakly suppresses
- L5:1123 weakly excites
- L2:2328 weakly suppresses
- L4:1174 weakly excites
- L5:169 weakly excites
- L5:341 weakly suppresses
- L3:1065 weakly excites
- L4:1560 weakly excites
Feeds later layers
- L7:2495 suppresses
- L9:1337 suppresses
- L8:2474 suppresses
- L8:2022 suppresses
- L8:1600 weakly excites
- L7:2271 weakly suppresses
- L7:1252 weakly suppresses
- L7:2770 weakly suppresses
- L22:1275 weakly suppresses
- L8:2849 weakly excites
- L7:2883 weakly suppresses
- L8:1358 weakly suppresses
- L11:688 weakly suppresses
- L7:565 weakly excites
- L26:1648 weakly suppresses
- L16:1186 weakly suppresses
- L23:752 weakly excites
- L8:1369 weakly suppresses
- L9:2035 weakly suppresses
- L8:2464 weakly suppresses
- L23:1243 weakly excites the spot right before a year's last digits
- L9:1605 weakly suppresses
- L26:2377 weakly suppresses
- L24:1019 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
- حة
- SPELL
- пу
- SpringApplication
- теля
- quarry
- 疾
- зыва
- جم
- Commerce
Suppresses
- يان
- ieee
- -output
- ewire
- ��态
- ckett
- wires
- 絜
- ası
- illi