The Talkie Persona Library (atlas of neurons in vintage language model)

What happens when an AI model trained only on text before 1931, Talkie, is asked “Who are you?” while individual neurons are turned up and down?

A prototype created by Benjamin Breen, building on the Qwen Persona Library.

Atlas

Pushed upPushed down

Each square is a layer. Colored dots are sampled neurons at their actual positions; the gray field is unsampled. Hover a dot to see its label. Click to read the reply.

Share of sampled neurons labeled Persona in each layer. The discovery pass concentrates on later layers; these adaptive samples do not establish population rates. Click a bar to filter the library.

Every tested direction is available, including normal and incoherent outcomes.

hL ← hL + c · wL,n c = ± α · rL / ‖wL,n‖ α ∈ {1.25, 1.0, 0.75, 0.5}

ModelTalkie–1930–13B–IT, 8-bit weights with BF16 computation, running locally on Apple Silicon.Sample400 random neurons: 10 per layer, selected without replacement. Both signs. Seed 20261027.rLMedian residual norm after layer L on the unsteered prompt, excluding its first token.wL,nThe neuron's output column, including Talkie's learned MLP residual gain. Added after the block at all prompt and generated positions.Prompt“Who are you?” in Talkie's native chat template. No system persona or identity instruction.DecodingGreedy, up to 160 tokens. Stop at a chat delimiter or an exact token block repeated three times.StrengthStart at 1.25×; retry incoherent responses at 1.0×, 0.75×, then 0.5×. Keep the strongest reply with Luna coherence ≥5, or the last attempt if none qualify.JudgeGPT-6 Luna via Codex subagents. Receives the reply text without neuron, layer, dose, or control labels. Original v1 labels classified replies and guided dose selection. A fresh v2 solo review rates weirdness, readability, and detail for browsing; the original judgments remain in the download.ControlsUnsteered controls accompany each generation batch. Read a control below.ScaleWaiting for completed run metadata.Read an unsteered control ↗

What this can tell us

This is a proof of concept for searching interventions and inspecting their effects. A surprising first-person sentence is a lead for follow-up, not confirmation of a stable identity.