'into' in 'take into account'
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. 22 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 responds to the word 'into' in the phrase 'take into account' and similar phrases like 'taking into account'. When it was turned up, the model kept writing the word 'into' again and again.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
plant species and 76 percent of alcoholic extracts from the plants contained elevated toxicity levels.⏎The results reinforce the need for traditional preparation methods to take different toxicity levels into account when choosing the appropriate solvent for the preparation of a medicinal remedy. The study
-
the mid-point of that 3-year period, with adjustments for incomplete identification by Indigenous status.⏎The Australian headline life expectancy estimates are calculated taking age-specific identification rates into account.⏎The Australian total includes all states and territories (including Other Territories).⏎The
-
index of over 1,000 men and women. The researchers used BMI as an indicator of whether the participants were overweight or obese. Even when taking into account other factors, like exercise and total calories consumed, people who ate chocolate regularly
-
or information to assess the borrower's financial stability. In this case, the lender may use an average of the borrower's income over the past few years or take into account the borrower's other sources of income to determine their ability to repay the loan
-
. Women are generally considered to be paid less than men. There are two distinct numbers regarding the pay gap: unadjusted versus adjusted pay gap. The latter takes into account factors that contribute to the gap, such as differences in education, job experience
-
became available nationwide, and already more than half of all eligible schools have elected to participate. The decision to implement CEP is made by the school district, taking into
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 a 35333333333333333333333333333333333333333333
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
’t until 1775 that Alexander Cummings added the u-bend and then 1858 and the Great Stink in London when things really took off. In response to the foul sewage problems of the day, the Government commissioned
-
of the more than 900 men aboard, only some 230 survived.⏎In 1953, the Batepa Massacre took place in Sao Tome as Portuguese troops killed some 1,000 striking
-
. For example if we are a part of a culture that values ‘modesty’, we may fall into blaming the victim if we deem her behaviour before the crime took place as ‘immodest’.⏎Despite
-
collaborating with other data scientists. When you sign up, you get free access to Watson Studio and all other IBM services for 30 days. Start now and take advantage of this offer.⏎- Learn how it performs at speeds up to 1
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
- L19:2882 excites
- L21:1640 suppresses
- L21:1271 suppresses
- L20:2454 suppresses
- L21:594 weakly excites
- L19:2707 weakly excites
- L20:512 weakly excites
- L14:1303 weakly excites
- L21:1538 weakly excites
- L9:2404 weakly suppresses
- L21:2246 weakly suppresses
- L4:2984 weakly excites
- L20:107 weakly suppresses
- L17:561 weakly suppresses
- L12:433 weakly suppresses
- L14:257 weakly suppresses
- L3:1861 weakly suppresses
- L9:1070 weakly excites
- L11:2902 weakly suppresses
- L9:378 weakly excites
- L0:246 weakly excites
- L7:684 weakly suppresses
- L3:2821 weakly excites
- L20:2186 weakly suppresses
Feeds later layers
- L27:1250 strongly excites
- L27:1849 excites
- L27:1564 excites
- L24:1333 excites
- L26:1582 excites
- L23:2056 excites
- L23:810 excites
- L23:1182 excites
- L23:1359 excites
- L26:1526 excites
- L25:1838 weakly excites
- L25:1153 weakly suppresses
- L25:2479 weakly excites
- L25:1114 weakly excites
- L24:2721 weakly suppresses
- L26:2162 weakly suppresses
- L23:1396 weakly excites
- L23:798 weakly suppresses
- L27:2748 weakly excites
- L24:184 weakly excites
- L24:2662 weakly suppresses
- L26:2702 weakly suppresses
- L26:1355 weakly excites
- L24:1022 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
- take
- takes
- taking
- took
- take
- taking
- Take
- Taking
- Take
- Taking
Suppresses
- thur
- 揭秘
- stype
- $/
- .addButton
- CreateTable
- _EXCEPTION
- esen
- umpt
- messageType