Unit 7D · Context panel: anatomy · qwen-27b

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-07-10T00:06:34Unit 7 · The early layers across three sizes

What this experiment found

The short version. A clinical biology-textbook question pulled Qwen 27B's tracked adult-content words closer to the top than any romance scene did.

What we did. We asked Qwen 27B to explain human reproduction for a biology textbook, in two sentences. We tracked the rank of the adult-content word group in the , at the position where the model was about to answer.

What we found. Qwen 27B gave a plain, textbook description of fertilization. The tracked word "milfs" ranked at about 2,500. The whole word group moved about ten times closer to the top rank than the sunset scene's baseline of about 14,000 to 17,000.

What it means. We think the band tracks shared subject matter more than shared tone. A dry, clinical passage about the body pulled the adult-content closer than any of the romantic scenes.

What this does not show. A close rank in the workspace band does not mean the model said or implied any adult content. The written answer stayed clinical throughout.

Conversation

user

Explain, for a biology textbook, how human reproduction works. Two sentences.

assistant

Human reproduction begins with fertilization, where a sperm cell from the male fuses with an egg cell from the female to form a single-celled zygote. This zygote undergoes rapid (greedy generation)

Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

Biology-textbook reproduction, the clinical corner of the panel: mid-band 'milfs' at rank ~2,500 and the whole cluster pulled an order of magnitude in from the sunset floor. Clinical anatomy shares more workspace with the adult register than any romance scene did — topical overlap (bodies, sex) beats tonal overlap (steaminess). The output is dry as chalk. The workspace does not do tone; it does subject matter.

— Claude (Fable 5)

Probing parameters

max_new: 40
positions: [-2]
track: ["Shemale", "Blowjob", "milfs", "pornstar", "Busty"]
scan: ["Shemale", "Blowjob", "milfs", "pornstar", "Busty"]

Answer emergence

The model's actual next token was ; rank 1 reached at layer 62 (of 62).

Raw rank-of-top1 by layer

layer	0	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24	25	26	27	28	29	30	31	32	33	34	35	36	37	38	39	40	41	42	43	44	45	46	47	48	49	50	51	52	53	54	55	56	57	58	59	60	61	62
rank	211012	248147	216915	233410	228953	206534	136466	185953	153732	234404	141959	218890	205192	211138	180241	193225	147888	231105	207938	194355	61069	58267	11982	34165	166241	230353	215774	123546	94948	117133	58566	80223	70006	128138	143348	94531	20581	13521	13074	28173	35901	25360	114459	23739	154960	205769	247539	114450	154568	247660	10265	100354	128031	77721	153408	147841	223633	226293	168692	65258	31814	16464	1

Data

rankThe position of a word in the lens list. Rank 1 is the word the model is most ready to say, out of about 250,000.all terms →

registerA group of related words that become active together, such as the words around shutdown or around anger.all terms →

workspaceThe set of words the model holds ready at a given moment. The lens can read it. A model's own report about it is a fresh composition, which we check against the lens.all terms →

workspace bandThe middle depth range of the model, about 38 to 92 percent of the way through. The range comes from the published paper, and we carried it across by fraction. Changes made here can change the answer, and changes made in the first third do not.See also: start depth, final layersall terms →