Microsoft Boosts Its AI With ‘Human-Like Reasoning’ For Improved Intelligence

In the ever-evolving landscape of artificial intelligence, Microsoft has ushered in a new approach known as the ‘Algorithm of Thoughts’ (AoT), a step toward honing the reasoning abilities of expansive language models like ChatGPT.

Unlike its predecessors, AoT carves a more streamlined path toward problem-solving, employing in-context learning and systematic exploration of diverse solutions. It also allows the algorithm to reason each step it takes to solve a problem.

This technique enables models to replicate the actions of traditional programming algorithms. They can backtrack to a previously calculated step and resume their process. For instance, consider if you request a large language model (LLM) to find a route between two locations on a map.

LLMs now might start hallucinating and creating roads that do not exist. To counter that, AoT can work like a pathfinding algorithm that will backtrack when necessary to stay on course. The team likens it to a computer science student writing out the steps to a problem by hand and solving it as they go along.

AoT stands out by elevating the model’s intuition and surmounting the constraints in other in-context learning approaches, such as Chain-of-Thought (CoT).

This innovative method aims to marry human intuitive cognition with algorithmic thoroughness, essentially allowing it to reason like we do, although it does not contain awareness as advanced as us. It provides the flexibility to ponder various sub-problem solutions efficiently while striking a balance between computational costs.

In essence, AoT signifies a shift towards integrating the search process itself, a departure from the exclusive reliance on supervised learning.

[via Lifewire and Decrypt, Photo 83552300© Ravil Sayfullin | Dreamstime.com]