How AI reads bunk | LLMsaurus

1 Your word measured

noun 9 senses 26 synonyms

Definitions come from WordNet, a hand-curated dictionary.

2 To an AI, it's pieces measured

DeepSeek V3 bunk 1 token

Mistral 7B bunk 1 token

Qwen3 8B bunk 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

bunker0.15 bunny0.14 busted0.13 downtime0.12 bedtime0.12 buck0.12 bourbon0.12 suck0.11

Dictionary synonyms measured

nonsense0.06 escape0.05 scat0.03 lam0.03 run0.02 rot0.02 beat0.02 feed bunk berth built in bed bunkum buncombe guff hogwash nonsensicality meaninglessness hokum bunk bed scarper turn tail run away hightail it head for the hills take to the woods

Shade shows how close the model puts each word to bunk. 7 of 26 dictionary synonyms are in this build; the rest have no vector to compare yet.

The model’s neighbours

bunker0.78 junk0.25 hunk0.18 chunk0.17 punk0.17 bed0.16 tank0.15 dorm0.14

Dictionary synonyms measured

nonsense0.09 run0.05 beat0.04 escape0.03 rot0.02 scat0.01 lam0.01 feed bunk berth built in bed bunkum buncombe guff hogwash nonsensicality meaninglessness hokum bunk bed scarper turn tail run away hightail it head for the hills take to the woods

Shade shows how close the model puts each word to bunk. 7 of 26 dictionary synonyms are in this build; the rest have no vector to compare yet.

The model’s neighbours

bunker0.90 hunk0.18 junk0.17 hunks0.16 bed0.16 bedtime0.15 bull0.14 dorm0.14

Dictionary synonyms measured

nonsense0.09 escape0.06 run0.06 rot0.04 beat0.04 scat0.03 lam0.01 feed bunk berth built in bed bunkum buncombe guff hogwash nonsensicality meaninglessness hokum bunk bed scarper turn tail run away hightail it head for the hills take to the woods

Shade shows how close the model puts each word to bunk. 7 of 26 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

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

wink0.112 barred0.112 trunk0.111 fucked0.110 bucks0.109

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

beds0.140 locker0.139 bust0.139 chunking0.137 rack0.135

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

tank0.142 cabin0.137 sink0.137 bogus0.134 bullock0.133

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

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