'see' in 'let's see' and 'see 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. 17 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 fires on the word "see" in phrases like "let's see" or "see if" or "see how". It is pushed negative by words about talking, such as "talked" and "speaks". Steering did not show a clear effect on the output.
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.
-
6 is conjectured to be the largest power of 2 not containing a zero. This simply stated conjecture has proven itself to be proof-resistant. Let us see why.⏎What is the probability that the nth power of two will not have any
-
- The longest beard ever was recorded at 17.4 feet long.⏎If you’re not already rocking a beard, you should try growing one out and see how you like it! However, if you don’t feel like having your own
-
. I'll see her tomorrow.⏎In #1 "going to" implies a previous arrangement or understanding
-
your kids can make a temporary lava lamp out of household items. Let’s see what you’ll need and how you can do it!⏎- Bottle (an empty
-
they recognise the parent’s voices. So reading to your baby for a few minutes gives them the comfort of your voice and listening abilities of the language.⏎Let’s see, as a parent how can we bring a change, which can make reading a
-
to as TBI, is thought to be extremely difficult to recover from. Well how about we inject some “female hormones” and see what happens?⏎In a recent article, Rita Rubin writes of an individual who suffered
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 need to not forget the need to not forget the need to not forget the need to not forget the need to not forget the need to not forget the need to not forget the need to not forget the need to not forget the need to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
IN, MS, NC, PA, RI, TN).⏎Yet again, these numbers underscore the importance of federal and state programs for low-income people. We’ve talked about this before,
-
ECSEL classroom loves implementing is one inspired by a TEDTalk by Louie Schwartzberg, a nature time-lapse photographer. In his TEDTalk, Schwartzberg speaks about his perspective as
-
about goat feed is “can goats eat strawberries”. There is much we can talk about this topic, hence, let’s spend some time reading the article and find
-
Graham Patrick manages this with a skilful blend of necessary detail and exciting descriptions of the possibilities of current and future organic synthesis.⏎Organic chemists like to talk in their own way using a peculiar language of structures and diagrams. While this is
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
- L16:1156 excites
- L5:1949 excites
- L10:1941 suppresses
- L10:388 weakly suppresses
- L16:973 weakly suppresses
- L4:2401 weakly suppresses
- L15:1688 weakly excites
- L16:1808 weakly suppresses
- L16:684 weakly excites
- L14:1692 weakly suppresses
- L10:2833 weakly excites
- L11:1687 weakly suppresses
- L13:1758 weakly suppresses
- L0:443 weakly suppresses
- L14:1024 weakly suppresses
- L5:448 weakly excites
- L15:2073 weakly excites
- L2:1378 weakly excites
- L15:2465 weakly suppresses
- L13:1254 weakly suppresses
- L14:611 weakly suppresses
- L4:2399 weakly suppresses
- L12:782 weakly excites
- L4:362 weakly suppresses
Feeds later layers
- L21:1176 weakly excites
- L19:1913 weakly excites
- L19:1177 weakly suppresses
- L20:766 weakly excites
- L23:542 weakly suppresses
- L27:1316 weakly excites
- L19:359 weakly excites
- L18:1489 weakly suppresses
- L23:33 weakly excites
- L19:2724 weakly excites
- L20:1015 weakly excites
- L20:2017 weakly excites
- L21:1016 weakly excites
- L20:2115 weakly excites
- L23:1387 weakly excites
- L22:1714 weakly excites
- L20:414 weakly suppresses
- L25:700 weakly suppresses
- L22:1085 weakly excites
- L22:2954 weakly excites
- L26:1605 weakly excites
- L25:2664 weakly excites
- L21:2117 weakly suppresses
- L19:945 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
- 'image
- 'o
- chner
- consulted
- ठ
- kn
- ken
- 的答案
- OPS
- 침
Suppresses
- 周
- anic
- Multiplicity
- 征
- anager
- width
- atz
- .prevent
- esi
- 件事