DLM Beyond LLM

• Why a shift in thinking is necessary

DLM is a model that avoids passive answers and creates active answers, which is a weakness of existing LLMs.

Artificial intelligence has evolved in the order of machine learning, deep learning, LSTM, Transformer, GPT, LLM, and LMM.

In the meantime, Korea has faithfully followed the development of artificial intelligence led by the United States. In this way, application technology can keep up, but it cannot act as the first penguin.

There are many paper mills around the world, but the first person to make paper is remembered as Chinese Chaerun or Papyrus. The first national branding, strategy, and technology reading are needed.

I’d like to suggest a DLM after LLM, although I’m free. DLM is a Discussable Language Model.

The problem with GPT is that users can only use it as much as they know it. In other words, you must ask questions and give commands well to answer at the user level.

This is a problem with the Generative Pretrained Transformer. In other words, if the questioner is a coding developer or major skilled in the major, he or she can receive an answer at that level. The weakness is that it can only be used at the level of the questioner. No matter how much knowledge you have in the world, it is difficult to go beyond the level of the questioner. Coding using GPT can lead to such regret. I want to make something in my imagination, but I don’t know how to specifically say such a command and ask a question, or how to present a prompt command.

Even though a doctor can give answers that are equivalent to 1,000 Ph.D.-level personnel, if the questioner is an elementary school student, the doctor should be able to make a product while teaching the elementary school student. That is the strength of DLM.

For example, let’s say I’m imagining a computer coding program. I don’t have the skills to implement it, I don’t have the a priori knowledge, and I don’t know how to make it, but can’t I make it in consultation with DLM?

When asked a question, GPT gives a wonderful answer. But sometimes it feels like an explanatory insect who talks about it again and again?

Imagine the use of DLM. For example, if you want to create a stock-related program, GPT4 has a long explanation. In fact, I have a rough idea of that explanation a long time ago… and I’m tired of it.

If you’re a DLM, you’ll ask like this. What kind of stock program? Hey, there are more than one or two stock programs. We have to advance the conversation like this.

Then the user doesn’t know the details, so I want to make a stock transaction that automatically earns profits. It is necessary to proceed with tiki-taka like this.

Even if I’m not an expert, I need to be able to dig into what the user wants even if the programmer has this or that ability to get counseling from a program expert.

There are people who learn through books, and people who learn through questions and answers. In order to better answer questions, DLM should ask users what they have not thought of.

Until now, LLM has been trained to answer users’ questions well. On the other hand, DLM should train users on asking questions and how to ask questions. In order to allow users and DLM to develop with each other in the discussion process, what do DLM users want, not just answering users’ questions, and what strengths and weaknesses, and consultations… Shouldn’t it be developed mainly on questions?

For the sake of the development of DLM, I think about starting with the field of theology for convenience. Theology is a study that has historically accumulated human questions and answers. There are many confrontations with each other, and there are many related papers, books, and sermons. I think DLM can be developed through theology, books, and papers.

We look forward to seeing a lot of related papers and ideas beyond GPT LLM to DLM.

DLLM #Discussable #Language #Model #LLM #GPT #Too #Question #Theology #Theological #Theological #Theological #Thesis