"over" in "take over"
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 "over" in the phrase "take over", as in "robots can take over rote tasks". It also shows a weaker negative response to "begin" in "To begin with". Turning it up gave unrelated text, so the steering did not confirm the pattern.
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.
-
. Here are some great examples in a musical format.⏎We will be looking at numbers next time then give and take over the next two weeks for Halloween, sobe sure to let me know by visiting
-
feel like they don’t know where they’re going and lose sight of their goal⏎- let procrastination take over⏎So how do you help your teen learn how to create an action plan when
-
which can be found on Facebook) have been working on the latter solution and have made significant progress towards the creation of an automated tool, potentially allowing computers to take over. But now,<|endoftext|>In today’s virtual workshop, artist Sophie Daly will show
-
, and rest are subject to a constellation of taboos and prescriptions. Sex is a no-no. Rest is mandated and traditionally facilitated by female relatives who take over errands and chores. Foods are divided into the approved (carrots, chicken
-
. Most automated machinery and tools can operate at an optimal level, as well as improve working conditions for those spending the most time inside such facilities. Robots can take over rote tasks, freeing up human workers to handle more exciting duties. They can
-
7-Separated from Baylor University.⏎1910-Moved to Fort Worth.⏎1918-College of Dentistry organized in Dallas by taking over the State Dental College which had been founded in 1905.⏎1
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 need to be more precise. I have to think about this problem in my own way. So, let me think again about the problem. The problem is to find the minimal value of the function. So, first, let me try
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
can the climate generation do?⏎To begin with, business as usual is not an option. These young people are green entrepreneurs
-
mediated mRNA isolation. Here, we describe a protocol that employs sample derived rRNA capture probes to remove rRNA from a metagenomic total RNA sample. To begin, both mosquito and microbial small and large subunit rRNA fragments are amplified from
-
to take a stab at it. Put on your seatbelt, because wrapping your head around TVOC can be a bit of a mental rollercoaster!⏎To begin our TVOC journey, let’s start off with a simple question: What is
-
advantages. Take a look.⏎Without a doubt, swept-source OCT has positively impacted and exceptionally contributed to
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
- L0:178 weakly suppresses
- L0:1301 weakly suppresses
- L1:1955 weakly excites
- L0:1224 weakly suppresses
- L0:2835 weakly excites
- L1:2907 weakly suppresses
- L1:3032 weakly suppresses
- L1:90 weakly excites
- L0:1962 weakly excites
- L0:659 weakly suppresses
- L1:1404 weakly suppresses
- L1:2318 weakly suppresses
- L1:379 weakly excites
- L1:1692 weakly excites
- L1:2197 weakly suppresses
- L1:2610 weakly excites
- L0:1204 weakly suppresses
- L1:2653 weakly suppresses
- L0:2348 weakly excites
- L1:1584 weakly suppresses
- L0:2952 weakly excites
- L1:559 weakly suppresses
- L0:48 weakly excites
- L0:1723 weakly excites
Feeds later layers
- L4:117 suppresses
- L9:2276 weakly excites
- L17:2499 weakly suppresses
- L14:1818 weakly excites
- L20:2721 weakly suppresses
- L3:1888 weakly excites
- L3:2293 weakly excites
- L14:2167 weakly suppresses
- L10:2664 weakly suppresses
- L9:1730 weakly suppresses
- L3:1661 weakly suppresses
- L3:1660 weakly excites 'ago' after a span of years
- L23:230 weakly suppresses
- L19:1224 weakly suppresses
- L4:2058 weakly suppresses
- L20:1995 weakly suppresses
- L4:825 weakly excites
- L10:459 weakly excites
- L7:1554 weakly suppresses
- L13:2964 weakly suppresses
- L20:2315 weakly suppresses
- L9:1919 weakly excites
- L4:360 weakly suppresses
- L12:2325 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
- _CN
- ariat
- 扇
- ult
- 丞
- ork
- 增量
- 价
- 侄
- enstein
Suppresses
- raud
- lesen
- aylight
- explanation
- материал
- .AI
- ASON
- truth
- :⏎⏎
- prites