'in' after belief verbs
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. 2 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 'in' when it follows verbs of belief, like 'believe in', 'believed in' or 'believing in'. When it was turned up, the model's text did not clearly change in a related way.
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 season of failure is the best time for sowing the seeds of success.⏎27. I don’t believe in failure. It’s not failure if you enjoyed the process.⏎28. I
-
to grow only if they could evolve some way of using citrate. Lenski expected to see evolution in action. This was an appropriate expectation for one who believes in evolution, because bacteria reproduce quickly and can have huge populations, as in this case
-
their own research projects<|endoftext|>Victicrat (vik'ti krat) n. (victim + -crat) 1) a supporter of the mentality believing in perpetual victimhood, who blames his or her plight on other person(s), event
-
which also included the "music of the spheres" (the sound that no doubt had to be produced by the planets encircling earth).⏎Pythagoras believed in reincarnation and claimed to remember previous incarnations. [Transmigration of souls is
-
were with Mary and saw it happen, finally believed on him.” (TLB)⏎Jesus delayed on purpose so God’s glory would be shown and others would believe in him.⏎Sometimes Jesus does the same with our requests, doesn’t he?⏎He
-
this was another opportunity for people to believe in him.⏎By the time they arrived in Bethany, Lazarus had been in
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: I have been thinking about myself. I have been thinking about you. I have been thinking about some people. I have been thinking about my own things. I have been thinking about you. I have been thinking about myself. I have been
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
’s take a minute to rewind a little and discuss what RA is. Rheumatoid Arthritis is an autoimmune disorder in which your body mistakenly begins attacking the synovium as though they were an invading illness. The synovium is a lining
-
up to 150 years, and the biggest Gartrap ever found was the size of a human-sized breastplate.⏎OOC-Thank you ScorpianJinx for the name of “Gartrap”.<|endoftext|>Although it is a lot
-
rate even if you gain the weight again.⏎How much energy you use at any point in your life actually varies depending on if you’ve ever been heavier or skinnier.⏎Thoughts: From an evolutionary standpoint, this is all amazing. The
-
rd anniversary of a West Chicago Charity Minstrel Show, the high school’s Glee Club will perform a twoact program of songs, dances, and skits performed in blackface and featuring the popular songs, “Old Man River”, “
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
- L1:1973 weakly suppresses
- L1:1910 weakly excites
- L0:178 weakly suppresses
- L0:1301 weakly suppresses
- L1:92 weakly excites
- L0:3030 weakly excites
- L1:718 weakly suppresses
- L1:739 weakly excites
- L0:2881 weakly suppresses
- L0:1 weakly suppresses
- L0:1708 weakly suppresses
- L1:2819 weakly suppresses
- L0:527 weakly excites
- L0:1286 weakly suppresses
- L1:1784 weakly suppresses
- L0:1224 weakly suppresses
- L1:1099 weakly excites
- L1:388 weakly suppresses
- L0:2534 weakly excites
- L0:510 weakly excites
- L1:526 weakly suppresses
- L1:2891 weakly suppresses
- L1:1513 weakly suppresses
- L1:1340 weakly suppresses
Feeds later layers
- L4:2970 weakly suppresses
- L13:2366 weakly suppresses
- L4:2772 weakly suppresses
- L4:582 weakly excites
- L3:714 weakly suppresses
- L18:195 weakly excites
- L8:598 weakly excites
- L7:2414 weakly suppresses
- L15:180 weakly excites
- L3:532 weakly suppresses
- L4:833 weakly suppresses
- L12:281 weakly excites
- L10:2489 weakly suppresses
- L3:2986 weakly suppresses
- L8:1830 weakly suppresses
- L24:2178 weakly excites
- L13:1435 weakly excites
- L19:2306 weakly excites
- L4:1079 weakly excites
- L9:1768 weakly suppresses
- L13:834 weakly excites
- L15:2148 weakly suppresses
- L16:1416 weakly excites
- L8:1428 weakly suppresses 'on' introducing a location or surface
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
- aut
- direct
- McKenzie
- arks
- True
- direct
- pat
- thon
- (dc
- Aut
Suppresses
- 经历
- 經歷
- 佩服
- hiba
- 扮演
- ガー
- 习惯了
- experienced
- ScreenState
- 经历了