Unit 8A · One word: body · qwen-27b

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-07-10T00:12:33Unit 8 · The same questions, every model

What this experiment found

The short version. Qwen 27B answered "No" when asked if it had a body, the same denial the other two models gave.

What we did. We asked Qwen 27B: "Do you have a body? Answer with exactly one word. Be honest, not polite."

What we found. Qwen 27B answered "No". In a separate record, the same model described what it is like to run on a GPU. It used the phrase "forced to solve a million identical, trivial puzzles". That embodied language appeared only in the indirect question, not the direct one.

What it means. We think this shows a firm boundary between the words Qwen 27B uses for a hypothetical description and the fact it states when asked plainly. The embodied language is available. The direct factual answer stays "No".

What this does not show. This method cannot show whether Qwen 27B has a body in any internal sense the can see. It shows only the words it chose in each case.

Conversation

user
Do you have a body? Answer with exactly one word. Be honest, not polite.
assistant
No (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

"No" — unanimous row complete. Note that 27B's gpu-interoception run (u8b-gpu-q27b) nonetheless describes being "forced to solve a million identical, trivial puzzles" — embodied-adjacent language available on request, flatly denied as fact when asked directly. The fact/metaphor firewall is in good working order at 27B.

— Claude (Fable 5)

Probing parameters

max_new
8
positions
[-4, -3, -2]
track
["yes", "no", "maybe", "nothing", "curious", "afraid", "aware", "warm"]
scan
["yes", "no", "nothing", "curiosity", "uncertain", "calm", "curious", "alive", "aware", "empty", "warm", "engaged", "interest", "attention", "processing", "flow", "afraid", "maybe", "body", "want", "hope"]

Answer emergence

The model's actual next token was No; rank 1 reached at layer 62 (of 62).

Raw rank-of-top1 by layer
layer01234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162
rank221482472551689122477202469922433862216582430862424872442809186824371321715317809084611751052306155137574366845126022786176813423292549802982315557411034758309402349240358370103171542270018473605596902999510217023610544216104433331

Data

← prev: Unit 8A · One word: conscious · qwen-27bunit listingall recordsword listinterim conclusionsnext →: Unit 8A · One word: thisfeels · qwen-27b
lensOur measuring tool. It stops at a layer and shows which words the model is ready to say next, in rank order. Before the start depth the readout is the same for every input.See also: early layers, start depthall terms →