"to" after "help" plus object
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. 1 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
The neuron fires on the word "to" in phrases like "help students to" or "helps you to", where someone is being helped to do something. Turning it up produced unrelated text, so the steering does not show this clearly.
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.
-
from accessing cattle water troughs, feed troughs and mineral licks, or entering farm buildings. Further information is available on the TB hub.⏎4. Help to roll out badger vaccinations⏎Badgers can be vaccinated against bTB. Bad
-
is involvement of immediate gain in academic career. It can also help you elevate your educational status and look good on a CV.⏎Honor classes can also help students to rise to the challenge of hard work, making them more poised and confident. Greater
-
can help the hospitality industry to cut costs and streamline interactions. One such example is the use of mobile electronic keys in hotels. This type of technology can help employees to check in and out of rooms more efficiently, which could mean reduced staffing costs and
-
lives or games that improve memory or numeracy skills.⏎3. Expand knowledge⏎One of the positive effects of smartphones is that they can help students to get information on different subjects very easily.⏎Smartphones are equipped with various applications,
-
part of an enzyme that is crucial for influenza virus replication.⏎The study was done on the 2009 pandemic influenza strain, but it will help scientists to design innovative drugs against all the different influenza strains, and potentially lead to a new
-
, or they may be hardware or software solutions. What matters primarily is that an ILS helps you to perform multiple tasks with the greatest ease in the least amount of time. Typical tools
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: this is the time I have been thinking about lately: this is the time I have been thinking about lately: this is the time I have been thinking about lately: this is the time I have been thinking about lately: this is the time
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Asia, and America. The American variety (var. Americana) is larger. Called⏎whistler, garrot, gowdy, pied widgeon, whiteside, curre, and doucker. Barrow's golden
-
, there are Polynesian figures who stole fire from the gods, and Loki, the Norse trickster god, is associated with fire. As Hans Biedermann notes in his The Wordsworth Dictionary of
-
which are open waters training, fitness and techniques. Also it is very important to protect the skin against the sun UVA and UVB radiation, and against Jellyfish and<|endoftext|>The moon is only two days past new today and is a waxing
-
-extinct frog from Sri Lanka; a highly venomous snakes from Australia; a fruit bat; a European mushroom; a jellyfish named after its victim, an American tourist who apparently died after being stung by
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
- L0:1301 suppresses
- L0:178 suppresses
- L0:1224 suppresses
- L0:83 weakly suppresses
- L0:1075 weakly excites
- L0:250 weakly suppresses
- L0:428 weakly suppresses
- L0:1993 weakly excites
- L0:370 weakly excites
- L0:1839 weakly excites
- L0:2092 weakly suppresses
- L0:191 weakly suppresses
- L0:36 weakly suppresses
- L0:3029 weakly excites
- L0:121 weakly excites
- L0:0 weakly excites
- L0:2712 weakly excites
- L0:992 weakly excites
- L0:89 weakly excites
- L0:1149 weakly excites
- L0:420 weakly excites
- L0:3010 weakly excites
- L0:896 weakly suppresses
- L0:1 weakly suppresses
Feeds later layers
- L14:785 weakly excites
- L15:874 weakly suppresses
- L17:1601 weakly suppresses
- L4:1012 weakly excites
- L16:1718 weakly excites
- L27:735 weakly excites
- L25:1243 weakly suppresses
- L12:564 weakly suppresses
- L23:1276 weakly suppresses
- L11:1993 weakly suppresses
- L2:2020 weakly excites
- L13:2211 weakly suppresses
- L4:823 weakly suppresses
- L6:998 weakly excites
- L8:2064 weakly suppresses
- L5:521 weakly suppresses
- L21:2504 weakly suppresses
- L8:917 weakly suppresses
- L20:906 weakly suppresses
- L11:3036 weakly excites
- L14:2635 weakly excites
- L2:2013 weakly suppresses
- L27:2932 weakly excites capital F starting a word
- L14:987 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
- �
- (js
- ما
- 拜
- 非凡
- �
- ��
- uset
- 不负
- 狄
Suppresses
- izziness
- ournal
- ária
- ários
- 理事长
- sheets
- oire
- écoute
- 却发现
- 院长