LLMs are A Dead End in Search for General Machine Intelligence: A Review

Abstract

This extensive review of large language models (LLMs) aims to highlight the importance of scaling the current generation of large language models toward artificial general intelligence, which is a dead end, while also considering the risks of unregulated use of such models. Through this, it is aimed to explicitly highlight the intelligence factor of current large language models and their malicious manipulative ability. While many large language model organisations compete to achieve better results by scaling up their models, this ultimately leads to the models collapse. It is too early to understand the development and benefits of large language models; many have cited LLMs as the primary means of achieving general intelligence agents. To counter this, this paper gathers and evaluates resources from multiple research articles and tests several frequently used LLMs, highlighting their importance in different scenarios. As these models are trained on a wide variety of data, they exhibit domain-independent intelligent behaviour but fail to exhibit causal intelligent behaviour.

Ähnliche Arbeiten