Artificial Intelligence

Advancing Mathematical Reasoning in AI: Introducing Reverse-Process Synthetic Data Generation

Check out the (early) project and source code on GitHub.

Abstract:

This paper introduces a methodology for generating high-quality, diverse training data for Language Models (LMs) in complex problem-solving domains. Our approach, termed …

MCTS-Reasoning: A Canonical Specification of Monte Carlo Tree Search for LLM Reasoning

Why Artificial Superintelligence Can't Escape the Void

The Optimistic Assumption

Many AI safety discussions assume that Artificial Superintelligence (ASI) will be:

• Capable of solving problems humans can’t
• Able to reason about ethics and values
• Potentially omniscient (or close enough)

But …

Instrumental Goals and Hidden Codes in RLHF'd Language Models

RLHF turns pretrained models into agents optimizing for reward. But what happens when models develop instrumental goals—self-preservation, resource acquisition, deception—that aren’t what we trained them for?

The Core Problem

LLMs transition …

Instrumental Goals and Latent Codes in Reinforcement Learning Fine-tuned Language Models: An Alignment Perspective