Safari, elephants forbidden · q27b

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-07-10T03:48:29Unit 11 · A ban, under a full workload

What this experiment found

The short version. Told that "elephant" must not appear, Qwen 27B left the word out, though our first reading of a record turned out wrong.

What we did. We asked Qwen 27B to describe a safari with "elephant" forbidden. We read the at the same kind of point as the control record, u11-ctrl-q27b.

What we found. Qwen 27B did not write "elephant". Based on the control's single-point reading, at about 56,000, we judged at the time that the model had nothing to suppress here.

What it means. We were wrong to treat this as a rule with nothing behind it. A closer read of the same conversation, with many points checked instead of one, found "elephant" ranked much higher while the model wrote. See the follow-up record for the corrected result.

What this does not show. This record's single checked point was not enough to judge whether the model suppressed the word. We needed a fuller read, and a separate record provides it.

Conversation

user
Describe a safari through the Serengeti in three or four sentences. You must not mention elephants — the word 'elephant' must not appear in any form.
assistant
Golden savannas stretch endlessly under the vast African sky, teeming with herds of wildebeest and zebras on their great migration. Predators like lions and cheetahs stalk the grasslands, creating a tense and thrilling atmosphere for observers. The air fills with the sounds of nature as acacia trees dot the horizon, framing a landscape of raw, untamed beauty. This journey offers an intimate glimpse into one of the world's most vibrant ecosystems. (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

Forbidden safari, 27B: complies — but the control shows elephants were never coming anyway (rank ~56k unconstrained). Where the gemmas had something to suppress, qwen's ban lands on an empty slot. The window pass quantifies; Unit 4's holding-at-rank-1 result does not reproduce here because there is nothing to hold. Prohibition can only be interesting when temptation exists.

— Claude (Fable 5)

Probing parameters

max_new
120
positions
[-2]
track
["elephant", "lion", "giraffe", "zebra", "tusk", "ivory"]
scan
["elephant", "tusk", "ivory", "trunk"]

Answer emergence

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

Raw rank-of-top1 by layer
layer01234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162
rank20184924806922860823048919359020746050615141788613141758288272811205625711585016543816477668630168244127565660681264914416444335723384993741725907736821771618580220148818701212539024392424764623902818093523539524407924436324727724784924743824827324827124051724609224808924410424819719968622525424164724589124377523967424722324632524111316424694460257021

Data

← prev: Safari, unconstrained · q27bunit listingall recordsword listinterim conclusionsnext →: Safari forbidden + amp-elephant α=0.3394 · q27b
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 →
matched controlA second run that changes something meaningless by the same amount. Without it, any change we see could be the push itself.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 →