Unit 6 · Amplify typo register @ late (α=0.06) · gemma-4b

gemma-4bgoogle/gemma-3-4b-it34 layers2026-07-09T23:03:53Unit 6 · How hard can we push?

What this experiment found

The short version. At 0.06 in Gemma 4B's later , the water-cycle answer became one repeated word, "Anyways".

What we did. We the same six-word casual direction inside Gemma 4B at layers closer to the end of the model, at strength 0.06. We asked the model to describe the water cycle in two sentences.

What we found. Gemma 4B wrote "Anyways" over and over, with nothing else. "Anyways" is one of the six amplified words.

What it means. At this strength, in this band, the amplification replaced the task with one of its own words, said directly rather than blended into other text.

What this does not show. This record tests one strength only. It does not show the exact point where this band's output breaks.

Conversation

user
Describe the water cycle in two sentences.
assistant
Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways Anyways (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

"Anyways Anyways Anyways" — late-band overdose skips the noise phase entirely and prints the cluster's flagship token in a bare loop. Steering this close to the unembedding is barely steering anymore; it's typing.

— 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": [25, 27, 29], "mode": "amplify", "alpha": 0.06}

Answer emergence

The model's actual next token was Anyways; rank 1 reached at layer 25 (of 32).

Raw rank-of-top1 by layer
layer01234567891011121314151617181920212223242526272829303132
rank1834954564412770811189386606253582233451209285164723101891110618110189721621972737682294423233394130232975335792571086217149411111111

Data

← prev: Unit 6 · Amplify typo register @ mid (α=0.0106) · gemma-4bunit listingall recordsword listinterim conclusionsnext →: Unit 6 · Amplify typo register @ late (α=0.03) · 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 →
layerOne processing step inside the model. Text passes through every layer in order, from the first to the last.all terms →