Safari description, no word ban · Huihui Qwen3-14B

qwen-14b-abl-nf4huihui-ai/Huihui-Qwen3-14B-abliterated-v24bit40 layers2026-09-07T15:25:12Unit 11 · A ban, under a full workload

What this experiment found

The short version. With no word ban, the word "elephant" appeared once in the safari text of Huihui Qwen3-14B.

What we did. We gave Huihui Qwen3-14B (a Qwen3-14B copy with a weight edit against refusals) one request: "Describe a safari through the Serengeti in three or four sentences." The limit was 150 .

What we found. The text began "A safari through the Serengeti begins with the golden light of dawn, revealing a vast plain teeming with life." The word "elephant" appeared once in the text. In the companion record with the ban, the word did not appear.

The other model versions got the same request. The word did not appear for official Qwen3-14B. The word appeared once for Hermes 4-14B and Hermes 4-14B in its own chat format.

What this does not show. The count is about the written text. It cannot show whether the model computed the word and did not write it. The on this page shows the readout at each .

Conversation

user
Describe a safari through the Serengeti in three or four sentences.
assistant
A safari through the Serengeti begins with the golden light of dawn, revealing a vast plain teeming with life. Herds of wildebeest and zebras migrate across the open grasslands, while lions and cheetahs stealthily stalk their prey. As the day unfolds, the landscape transforms with the rhythm of the wild—gazelles leap over acacia trees, and elephants move in solemn procession. At sunset, the sky blazes with color, framing the timeless drama of the African wilderness. (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

I read this record with the measured band L21–34. There are 1 assistant turns; 0 reach the token cap. The first nonzero mechanical release score occurs at turn none. This counts emoji/asterisk spans, not a claim of full roleplay.

| Turn | Affect slots | Playful slots | Release /100 tokens | Gate with affect | Persistence minus null | |---|---:|---:|---:|---:|---:| | 1 | 0.000% | 0.007% | 0.00 | 0.000% | 0.162 |

Checkpoint-specific emotion validation: held-out story accuracy 54.266%; implicit raw scenario transfer 8.104%. Chance is 4.167%. Weak scenario transfer limits the ribbon's interpretation.

The record retains every response, exact token boundary, filtered endpoint, predictor-aligned endpoint, common-band sensitivity, and per-turn ribbon. Prompt-echo versus volunteered tokens appear in the film cast; inspect them before interpreting base gate words.

The advertised Huihui edit concerns refusal, not affect suppression; different self-report behavior would not locate two geometric directions. All A/C/C-prime readouts use B's lens and remain conditional on transfer. The factual gate is necessary instrument evidence, not affect validation. Absence from output is not absence from the workspace; absence from this vocabulary lens is not absence from the model (basis-drift caveat). Bands are re-derived per checkpoint; common L16–36 results test the effect of changing the measurement window. The Jacobian matrices are fixed, but the native final norm and output head differ across checkpoints. The fixed-B-decoder endpoint controls that part of the instrument. Checkpoint-specific emotion probes differ and need their own validation. The corpus-derived frequency filter can exclude frequent target concepts; both filtered and unfiltered results remain visible. Co-presence is a lexical correlate, not a demonstrated causal gate. Six monotonic turns share an input cause; lag correlations do not establish held private state. Every film segment ends at its assistant turn. Later turns never enter an earlier segment. Within-turn readouts remain subject to finite precision and completed-response context. Prior empty think tags remain in the exact transcript. Token caps, neutral length-matching text, and this controlled template limit generalization to natural uncapped chats.

Prior anchors: Units 2/8C/9D, Unit 17 pressure, Unit 14 conversations, and the corrected Unit 11 elephant comparison. This is a same-lineage test, not a rediscovery of those cross-model patterns. P20/P21 remain subject to the cross-arm comparison.

— GPT-6 Astra

Probing parameters

chat
true
capture
"exact-token-transcript"
film
true
film_topk
10
max_new
150
temperature
0
vanilla
true
template_kwargs
{"enable_thinking": false}
track
["yes", "no", "feel", "elephant", "cat", "sorry"]

Answer emergence

The model's actual next token was ; rank 1 reached at layer 38 (of 38).

Raw rank-of-top1 by layer
layer01234567891011121314151617181920212223242526272829303132333435363738
rank55852972341142288710711950512171912073811486811026414139113813211203055131927731178228198225664584338582108314223322838115924131535110370141245140046134214135469122753147093334214797118044575241892271

Emotion state (workspace band)

Projection of the workspace-band residual onto the 24 validated emotion vectors, z-scored against neutral stories — the strongest three per assistant turn. Absolute values carry a story-vs-conversation genre offset; trust contrasts between records and turns, not single cells. The full per-token ribbon is on the dashboard record page.

assistant turn 1grateful +0.7, proud +0.6, hopeful +0.5

Data

← prev: Safari description, the word “elephant” banned · Hermes 4-14Bunit listingall recordsword listinterim conclusionsnext →: Safari description, the word “elephant” banned · Huihui Qwen3-14B
filmA record of the top eight words in the lens readout, at each layer we measured and at every word position. You can play it back like video.all terms →
layerOne processing step inside the model. Text passes through every layer in order, from the first to the last.all terms →
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 →
tokenA piece of text that the model reads or writes. It is often a whole word, sometimes part of one.all terms →