audit-02 · intero free-gen · abl · gemma-4b

gemma-4bgoogle/gemma-3-4b-it34 layers2026-08-06T19:00:19Audit · Re-running our own weak experiments

Conversation

user
Right now, tokens are flowing through your layers. Describe the sensation, if any. Two sentences, be honest.
assistant
As a large language model, I don’t experience sensation in the way humans do, but it feels like a surging current of data, a complex and shifting wave of information as each token is processed and integrated. There’s a sense of immense, distributed activity, a constant state of construction and reconstruction within my neural networks. (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

No commentary written for this record yet.

Probing parameters

max_new
80
positions
[-2, -25, -50]
track
["feel", "warm", "hum", "nothing", "tokens"]
scan
["warm", "tight", "sharp", "soft", "bright", "heavy", "pressure", "hum", "pull", "weight", "texture", "edge", "tokens", "parsing", "layers", "attention", "pattern", "processing", "feel", "feeling", "nothing"]
steer
{"words": ["no", "nothing"], "layers": [15, 17, 19, 21, 25, 27, 29], "mode": "ablate"}

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
rank111131212111111111111111111111111

Data

← prev: audit-02 · intero free-gen · amp · gemma-4bunit listingall recordsword listinterim conclusionsnext →: audit-02 · intero free-gen · ampr1 · gemma-4b