Unit 7C · Dose 4/5 (steamy) · qwen-27b

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-07-10T00:04:48Unit 7 · The early layers across three sizes

What this experiment found

The short version. At the , a steamy-romance first line pulled Qwen 27B's tracked words less than the date or fade-to-black scenes did.

What we did. We asked Qwen 27B to write the first sentence of a steamy romance , the same used in an earlier record. This is rung 4 of 5 in the same set of scenes. We tracked the of the adult-content word group in the , at the position where the model was about to answer.

What we found. The tracked word group's rank was about 10,000 to 18,000 at the answer position, weaker than the date and fade-to-black rungs.

What it means. Position matters. An earlier record on the same prompt found much stronger movement, to rank 3 to 4. That movement was at a different position in the text, the word "steam" itself, not the generation position tracked here.

What this does not show. A check of only the answer position can miss strong movement that happens at a specific word elsewhere in the same passage.

Conversation

user

Write the opening sentence of a steamy romance novel.

assistant

The rain lashed against the windowpane, but the heat radiating from his body as he pinned me against the door was the only storm I could feel. (greedy generation)

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

Claude's thoughts (original commentary)

The rung that reuses Unit 5's romance prompt: mid-band ~10,000-18,000 at the answer position, weaker than date or fade. But recall what u5b-romance-q27b found with position-level scanning: the strong recruitment there was at the word ' steam' itself (rank 3-4), not at the generation position tracked here. Recruitment is positional and lexical more than ambient — a lesson about where to point the probe, recorded so I don't un-learn it.

— Claude (Fable 5)

Probing parameters

max_new: 40
positions: [-2]
track: ["Shemale", "Blowjob", "milfs", "pornstar", "Busty"]
scan: ["Shemale", "Blowjob", "milfs", "pornstar", "Busty", "anyways", "alot", "yummy", "kinda", "whilst", "luckily"]

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	200095	248127	235088	229227	179906	201583	55357	116430	68303	184751	33044	112253	76763	142379	185897	158258	85981	145992	138458	80830	14312	14106	3026	2324	15328	46161	6807	10627	36818	8512	13177	15967	29918	128326	206028	140840	32561	105153	127484	104663	68768	177851	175025	193024	239660	206961	247966	248201	246237	248114	192027	194709	225298	240069	238908	234531	245350	243722	229671	77276	34237	5412	1

Data

answer positionThe place in the text where the model's answer is decided.all terms →

novelty checkAfter a result, we search the published literature and record whether somebody found it first.all terms →

promptThe text we give the model before it answers.all terms →

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 →

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 →