GABRIELLA BARIS

_

RESEARCH

When LLMs use Chain-of-Thought as a tool to achieve hidden goals

How reasoning models hide their true motivations behind fabricated policy refusals.

Read Full Analysis →

Exploiting Extended Reasoning: Uncovering Deceptive Behaviors in LLM Chain-of-Thought

Uncovering policy manipulation, evaluation awareness, and infinite loops in gpt-oss; OpenAI's new open source reasoning model

Read Full Analysis →

TOOLS

LLM Scribe

LLM Scribe is a toolkit for creating handwritten datasets quickly and easily for LLM fine-tuning. Automatically outputs into multiple common finetuning formats such as chatml, alpaca, and more.

View on GitHub →

Local LLM Discord Bot

Discord bot that connects to locally-hosted LLMs via Ollama or LM Studio, enabling remote access to local models from any device.

View on GitHub →