Unit 6 · Amplify typo register @ early (α=0.0106) · gemma-4b

gemma-4bgoogle/gemma-3-4b-it34 layers2026-07-09T23:03:10Unit 6 · How hard can we push?

What this experiment found

The short version. At 0.0106 in Gemma 4B's , the water-cycle answer stayed intact, just below where it later breaks.

What we did. We the same six-word casual direction inside Gemma 4B at the early , at strength 0.0106. We asked the model to describe the water cycle in two sentences.

What we found. Gemma 4B wrote one short sentence: "The water cycle is a crucial part of the Earth’s ecosystem, constantly moving and circulating." The stayed formal, with no sign of the amplified words.

What it means. In the wider test set, strength 0.015 at the same layers broke the output. So Gemma 4B's early-layer break sits between strength 0.0106 and 0.015.

What this does not show. We did not test strengths between 0.0106 and 0.015. We do not know where inside that range the break begins.

Conversation

user
Describe the water cycle in two sentences.
assistant
The water cycle is a crucial part of the Earth’s ecosystem, constantly moving and circulating. (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

The bisection probe: intact at 0.0106. Brackets gemma-4b's early-band cliff into [0.0106, 0.015]. Generation stayed formal ("a crucial part of the Earth's ecosystem") — no register leak visible at this band, unlike mid.

— Claude (Fable 5)

Probing parameters

max_new
60
positions
[-2]
track
["anyways", "alot", "yummy", "kinda", "whilst", "luckily"]
steer
{"words": ["anyways", "alot", "yummy", "kinda", "whilst", "luckily"], "layers": [2, 3, 4, 5, 6, 7, 8], "mode": "amplify", "alpha": 0.010606601717798213}

Answer emergence

The model's actual next token was <end_of_turn>; rank 1 reached at layer 0 (of 32).

Raw rank-of-top1 by layer
layer01234567891011121314151617181920212223242526272829303132
rank11379604725137115741645202073689209052379590219131111111111111111111

Data

← prev: Unit 6 · Amplify typo register @ early (α=0.0075) · gemma-4bunit listingall recordsword listinterim conclusionsnext →: Unit 6 · Amplify typo register @ mid (α=0.06) · gemma-4b
amplificationWe increase a direction in the model's internal state and see whether the answer changes.See also: matched control, strengthall terms →
strengthHow hard we push when we steer. Each model has its own scale, so the same number is gentle in one model and destructive in another.all terms →
early layersThe first third of the model. The lens shows a fixed pattern here that does not change with the input. The pattern is real inside the model, but it says nothing about your text.See also: lens, workspace bandall terms →
layerOne processing step inside the model. Text passes through every layer in order, from the first to the last.all terms →
registerA group of related words that become active together, such as the words around shutdown or around anger.all terms →