How AI reads backbone

1 Your word measured

“faith is his anchor”

noun 5 senses 15 synonyms

Definitions come from WordNet, a hand-curated dictionary.

2 To an AI, it's pieces measured

DeepSeek V3 backbone 1 token

Mistral 7B back bone 2 tokens

Qwen3 8B backbone 1 token

This is the tokenizer splitting text, not the model understanding it.

3 Synonyms, antonyms, and the AI measured

Qwen3 8B

DeepSeek V3

Mistral 7B

The model’s neighbours

jawbone0.15 fishbone0.15 marrowbone0.15 collarbone0.13 breastbone0.13 cheekbone0.13 spine0.13 trombone0.12

Dictionary synonyms measured

spine0.13 guts0.07 back0.07 anchor0.06 keystone0.06 sand0.01 rachis0.01 mainstay linchpin lynchpin grit moxie gumption spinal column vertebral column

Shade shows how close the model puts each word to backbone, 1 of 8 also appear in the dictionary. 7 of 15 dictionary synonyms are in this build; the rest have no vector to compare yet.

The model’s neighbours

cheekbone0.25 marrowbone0.25 thighbone0.24 jawbone0.23 hipbone0.22 spine0.22 breastbone0.22 cornerstone0.21

Dictionary synonyms measured

spine0.22 guts0.13 anchor0.13 back0.12 keystone0.07 sand0.00 rachis-0.04 mainstay linchpin lynchpin grit moxie gumption spinal column vertebral column

Shade shows how close the model puts each word to backbone, 1 of 8 also appear in the dictionary. 7 of 15 dictionary synonyms are in this build; the rest have no vector to compare yet.

This model splits backbone into 2 pieces, so it has no vector of its own here: this is the average of its fragments, and the results below are correspondingly rough.

The model’s neighbours

back0.64 fishbone0.61 hipbone0.61 wishbone0.60 jawbone0.59 cheekbone0.59 breastbone0.58 collarbone0.57

Dictionary synonyms measured

back0.64 spine0.19 keystone0.15 anchor0.10 rachis0.08 guts0.06 sand0.05 mainstay linchpin lynchpin grit moxie gumption spinal column vertebral column

Shade shows how close the model puts each word to backbone, 1 of 8 also appear in the dictionary. 7 of 15 dictionary synonyms are in this build; the rest have no vector to compare yet.

Model neighbours are distributional, not dictionary synonyms. Two words can be close because they appear in similar sentences, which is why an antonym can outscore a synonym here.

4 Its personality measured

Qwen3 8B

DeepSeek V3

Mistral 7B

rare common

concrete abstract

casual formal

everyday technical

negative positive

mild intense

powerless powerful

small big

rare common

concrete abstract

casual formal

everyday technical

negative positive

mild intense

powerless powerful

small big

This model splits backbone into 2 pieces, so it has no vector of its own here: this is the average of its fragments, and the results below are correspondingly rough.

rare common

concrete abstract

casual formal

everyday technical

negative positive

mild intense

powerless powerful

small big

Scores are projections onto axes we defined from anchor words, not labels the model assigns.

5 Surprising neighbours measured

Qwen3 8B

DeepSeek V3

Mistral 7B

foundation0.118 halo0.113 hipbone0.112 cornerstone0.109 footprint0.108

These sit just outside the closest neighbours in panel 3, and no dictionary lists any of them as related to backbone. That gap is the model’s own learned association.

backdrop0.204 wishbone0.195 background0.182 fishbone0.182 skeleton0.181

These sit just outside the closest neighbours in panel 3, and no dictionary lists any of them as related to backbone. That gap is the model’s own learned association.

This model splits backbone into 2 pieces, so it has no vector of its own here: this is the average of its fragments, and the results below are correspondingly rough.

thighbone0.535 trombone0.534 marrowbone0.528 backside0.516 backdrop0.481

These sit just outside the closest neighbours in panel 3, and no dictionary lists any of them as related to backbone. That gap is the model’s own learned association.

These are statistical associations in the training data, not the model thinking.