"to" after begin or start
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. 4 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 lights up on the word "to" right after verbs like "begin," "began," or "start," as in "we begin to collect data." Its strongest negative cases are unrelated, such as line breaks in lecture transcripts. Turning it up did not produce a clear change in the model's replies.
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.
-
, and once we have a proof of concept that we can build a device that has these capabilities in principle, we can go to the testing stage where we begin to collect data from real epilepsy patients.⏎“Provid<|endoftext|>Dashes vs. Unders
-
puberty. “Many parents don’t want men to teach their daughters,” said Pushpa Kumar, the Okhla’s center’s teacher.<|endoftext|>When we begin to talk about "deferred" items in relation to accounting transactions, we begin to
-
by the Council of Chief State School Officers (CCSSO) and the National Governors Association (NGA).” My question is this: When did Montana begin to be represented by associations and councils rather than our elected representatives in the legislature? Denise
-
Performance measures for primary prevention of cardiovascular disease in adults: A report of the American College of Cardiology Foundation<|endoftext|>One of the interesting things that happens when one begins to coach others is that the skill being taught becomes embedded more deeply into one’s own
-
foresters talk about planting oaks throughout the state [and] having a forest products industry that begins to move toward oaks. If these come true, that's going to change the
-
is in those moments they become reenergize for and reinvested in further practice.⏎It is with this notion of practice in my head that I began to explore what practice is in education. How it is like practice in sport? I
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 I can do this problem, and then I can use the knowledge that I have. It's a problem about how I can do this problem, and then I can use the knowledge that I have. It's a problem about how I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
few common examples.…As I discussed in the 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
-
.…It's less powerful than even a small…watch battery.…For processing and mixing,…the weak mic level signal…needs a microphone preamplifier,…or mic preamp, to boost it
-
is useful when we do our u-substitution,⏎- we could say that du is equal to 5 cosine 5x.⏎- Now when you look over here, we don't have quite du there. We have just cosine of 5
-
Candles are made from soybeans, a natural, renewable source grown in many regions by local farmers....right here in the USA! A candle made from soy wax burns cleaner, lasts longer and
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
- L3:191 suppresses
- L3:213 suppresses
- L3:59 excites
- L2:493 suppresses
- L1:2830 weakly suppresses
- L1:2459 weakly excites
- L3:456 weakly suppresses
- L1:939 weakly suppresses
- L3:179 weakly excites
- L2:111 weakly suppresses
- L1:2658 weakly suppresses
- L1:1464 weakly excites
- L0:178 weakly suppresses
- L3:1311 weakly suppresses
- L1:74 weakly excites
- L3:2369 weakly suppresses
- L2:70 weakly excites
- L0:1224 weakly suppresses
- L2:2116 weakly excites
- L2:1036 weakly excites
- L2:1599 weakly excites
- L0:1301 weakly suppresses
- L3:50 weakly excites
- L1:23 weakly suppresses
Feeds later layers
- L9:776 suppresses
- L9:2779 suppresses
- L6:49 excites
- L5:1047 suppresses
- L20:2621 suppresses
- L9:1626 suppresses
- L23:65 excites
- L8:874 weakly suppresses
- L16:123 weakly excites
- L18:1766 weakly suppresses
- L9:1149 weakly suppresses
- L10:1464 weakly excites
- L5:2325 weakly suppresses
- L16:2405 weakly suppresses
- L12:647 weakly excites
- L12:2218 weakly suppresses
- L7:1700 weakly suppresses
- L7:1979 weakly excites
- L16:89 weakly excites
- L9:1533 weakly suppresses
- L9:627 weakly excites
- L22:5 weakly excites
- L7:1843 weakly excites
- L15:73 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
- prints
- 锰
- stands
- -fin
- 受限
- TEMPL
- kdir
- leaf
- 름
- uced
Suppresses
- IPA
- osemite
- ALA
- Vimeo
- ynchronize
- quelle
- gli
- translating
- .translate
- iota