psyctl: Steer LLM Personality Without Fine-Tuning

rick — Sun, 15 Mar 2026 13:03:46 +0000

What if you could make an LLM more extroverted — without any training?

That's the idea behind psyctl, a CLI tool I'm building at Modulabs Persona Lab. It lets you extract personality vectors from a model's internal activations and inject them during inference to shift behavior. No fine-tuning, no LoRA, no RLHF — just vector addition.

How It Works

The technique is called Contrastive Activation Addition (CAA). Here's the pipeline:

Generate a contrastive dataset — pairs of responses that differ only in personality (e.g., extroverted vs. neutral)
Extract a steering vector — compute the mean activation difference between the two response sets
Inject the vector at inference — add the vector to a target layer's activations during forward pass
Validate with psychological tests — run standardized inventories to measure the personality shift

What's fascinating is that meaningful behavior changes emerge from simple vector arithmetic on activations — no gradient updates needed.

The CLI

psyctl automates the entire pipeline:

# Generate contrastive personality dataset
psyctl dataset.build.steer --personality Extroversion --output ./data

# Extract steering vector using mean difference method
psyctl extract.steering --dataset ./data --method mean_diff --output ./vec.safetensors

# Apply steering and generate text
psyctl steering --steering-vector ./vec.safetensors --input "Tell me about yourself"

# Validate with psychological inventory
psyctl benchmark inventory --steering-vector ./vec.safetensors

Extraction Methods

Two approaches are supported:

Mean Difference — a statistics-based method that computes the mean activation difference between positive and neutral responses. Fast and simple.
BiPO (Bidirectional Preference Optimization) — an optimization-based method using DPO loss to learn a more refined steering direction.

Evaluation

How do you measure if an LLM's personality actually changed? With the same tools psychologists use on humans:

IPIP-NEO — measures the Big Five personality traits (Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism)
NPI-40 — measures narcissistic personality traits
MACH-IV — measures Machiavellianism

psyctl administers these inventories automatically and compares scores before and after steering.

Compatibility

Works with HuggingFace Transformers models including:

Llama 3.x
Gemma 3
Qwen 2.5
Mistral

Any decoder-only transformer with accessible intermediate layers should work.

Key Papers

The implementation builds on these research papers:

runpod-log: A CLI Tool for Viewing RunPod GPU Pod Logs

rick — Sun, 15 Mar 2026 12:57:05 +0000

Why I Built This

RunPod is great for spinning up GPU pods, but there's one frustrating gap: you can't view pod logs from the CLI.

The official runpodctl lets you start, stop, and manage pods — but if you want to check logs, you're forced to open the web console every time. This is especially painful when you're managing multiple pods or trying to build automation scripts.

So I built runpod-log, a simple CLI tool that fetches RunPod GPU pod logs directly in your terminal.

Features

Fetch logs instantly — get both container logs and system logs in one command
Real-time monitoring — stream logs to a file with the tail command
Automatic authentication — browser-based login via Playwright with automatic JWT token refresh

Installation

pip install runpod-log

Usage

# Login (opens browser for authentication)
runpod-log login

# Fetch logs for a specific pod
runpod-log logs <pod-id>

# Filter by log type
runpod-log logs <pod-id> --only container
runpod-log logs <pod-id> --only system

# Real-time log monitoring
runpod-log tail <pod-id> ./logs.txt

# Logout
runpod-log logout

How It Works

Authentication

Since RunPod doesn't offer a public API for logs, runpod-log uses a browser-based authentication flow:

runpod-log login opens a Playwright browser window
You log in to RunPod normally
The tool captures the JWT token from hapi.runpod.net requests
Session data is stored locally — no repeated logins needed

Token Refresh

JWT tokens expire after about 60 seconds. When that happens, runpod-log automatically spins up a headless browser to fetch fresh credentials — no manual intervention required.

Log Retrieval

Under the hood, it calls the undocumented https://hapi.runpod.net/v1/pod/{pod_id}/logs endpoint to fetch both container and system logs.

Why Not Just Use the Web Console?

Automation: pipe logs into other tools, grep for errors, trigger alerts
Multi-pod management: check logs across pods without switching browser tabs
SSH workflows: view logs on remote machines without a GUI

DEV Community: rick

psyctl: Steer LLM Personality Without Fine-Tuning

What if you could make an LLM more extroverted — without any training?

How It Works

The CLI

Extraction Methods

Evaluation

Compatibility

Key Papers

Links

runpod-log: A CLI Tool for Viewing RunPod GPU Pod Logs

Why I Built This

Features

Installation

Usage

How It Works

Authentication

Token Refresh

Log Retrieval

Why Not Just Use the Web Console?

Links