| AI AGENTS| LLMs| ARTIFICIAL INTELLIGENCE|
“More human than human” is our motto. — Blade Runner
Agents are one of the most interesting applications for LLMs. An agent AI would make it possible to be able to make life easier for humans. But how do we evaluate their abilities? How can we be sure that they can interact with the real world?
We discuss this in this article
LLMs have brought about a revolution; scaling of models has made it possible to solve tasks with a quality never seen before. Models such as BERT and ChatGPT scaled the benchmarks for tasks such as text classification, information extraction, and question answering. These tasks, however, have one limitation-they are all text tasks. The model learns from the text but interacts only with the text.
On the one hand, these tasks are complex and not easy to solve, and the world is not only made of text. That is why the concept of AI agent was also discussed:
Typically in AI, an agent refers to an artificial entity capable of perceiving its surroundings using sensors, making decisions, and then taking actions in response using actuators. (source)
The development of LLMs has accelerated interest in these AI agents. LLMs have demonstrated particular capabilities for “knowledge acquisition, instruction comprehension, generalization, planning, and reasoning, while displaying effective natural language interactions with humans.” Therefore, it was thought that they could be used as a focus for agent development.