Generative AI Gets Shaken Up By Next-Concept Modeling That Produces Concepts First And Words Second

In today’s column, I examine a newly devised method to generate text that smartly reworks the inner functionality of generative AI and large language models (LLMs). This innovative approach could give conventional LLMs quite a run for their money. You see, the usual or common approach employed by nearly all popular LLMs such as ChatGPT, GPT-5, Claude, Grok, Gemini, and Copilot is to generate a response on a word-by-word basis, also known as token-by-token generation. You can envision this as laying bricks one at a time. Each brick is placed before the next brick is added to the construction.

Instead, in a brand-new approach coined next-concept modeling (NCM), the AI first starts by broadly identifying the words and concepts that will be the overarching skeleton or framework for the text that will ultimately be generated. A series of successive passes deepens the portrayal by incrementally filling in the words. You might compare this to writing a story or composing a response by doing a kind of broad-strokes version that gets progressively detailed and polished. There are advantages to this coarse-to-fine method of NCM, including leaning into parallelism to speed up the generation process and potentially reducing the total number of steps toward producing responses.

Let’s talk about it. This analysis of an innovative AI breakthrough is part of my ongoing Forbes column coverage on the latest in AI, including identifying and explaining various impactful AI complexities (see the link here).

How Things Usually Work

Text-Based Diffusion As An Alternative

Third And Newest Method Comes To The Fore

In recap, there is the typical token-by-token approach for LLMs, and there is also the creative but rarely used text-based diffusion approach. We can add a third method to the search for additional ways to generate text. A newly devised approach known as next-concept modeling (NCM) offers an innovative scheme.

Research On NCM

Thinking Outside The Box

A few final thoughts for now.