[Preview of the series]

① A conceptual dictionary of intermediate difficulty disguised as a beginner.  (→ we are here now!)

② Generative AI of the future & The Future for Generative AI

It's an intermediate level of difficulty disguised as a beginner. You'll be able to bring useful information that's easy to read but unexpected. Predicting various questions and clear answers about Generative AI and the Generative AI of the near future, we've created an easy, friendly Q&A about what to do with alertness, looking at and using AI. I'll tell you clearly what you know, and clearly what you're confused about.

Interview with ChatGPT

Why does ChatGPT speak from a first person perspective?

The reason ChatGPT, the current LLM frontrunner, spoke from a first-person perspective was because it was a design decision to interact more naturally and intimately with users. It was set up to use "first-person pronouns" as an easy-to-communicate conversation.

Which question did ChatGPT hear the most from Korean users?
Korean users asked the most questions, "How can I study more efficiently?" ChatGPT added, "Especially by test takers or college students."

What is ChatGPT ranked in the minds of users as a search tool?

According to Open Survey's Search Trend Report 2024, which was released in March, ChatGPT utilization rate was 17.8%, ranking eighth as a search tool. Naver (87%) was the undisputed No. 1 ranking. It seems that it has not gained much trust as a search tool yet.

What is ChatGPT's recognition vs usability vs reliability?

According to the 'Search Trend Report 2024' survey, ChatGPT's awareness was very high at 80.8 percent. On the other hand, only 34.5 percent had actual experience using it. It's amazing that it's less than half the usage rate compared to recognition. In the reliability evaluation of ChatGPT answers, only 40 percent of users said they were "reliable." There needs to be a lot of improvement in terms of the reliability of the answers.

Generative AI dictionary

1) Large Language Models (LLM)

⁕ Definition: A Generative AI system that generates answers to questions or instructions in common language.
⁕ Easy but deep explanation: Using a special multi-layered and multi-faceted neural network called 'Transformer', we learn a vast amount of natural language data collected from the Internet. After you complete basic learning, you go through the socialization stage through reinforcement learning through human feedback. It can be said that it is a work of receiving education that does not discuss prohibited content such as 'how to make a bomb' or 'how to avoid the law'.

2) Embedding

⁕ Definition: a representation of natural language written by a person as a numerical vector (list) that can be understood by a machine.
⁕ Easy but deep description: LLM transforms each word into a specific type of numeric vector called embedding. Words with similar meanings are represented by similar vectors. For example, words like 'house,' 'apartment,' and 'residence' are represented by similar vectors. Word embedding creates a unique vocabulary list, expressed as a statistical measure of word association. Humans cannot read it, but computer programs can.

3) Transformer¹⁾

⁕ Definition: A special type of neural network used by LLM.
⁕ Easy but deep explanation: When we leave the house, we hear a variety of noises at once. You can hear a lot of sounds, including the sound of cars, birds, children's voices, digging in construction sites, sirens, and sneezing by passing passers-by. Transformers do the same, just as we collect important information by paying particular attention to some of these sounds. The structure of the sentence and the intention of the question are captured by weighing the gravity of each word in the input sentence. When you ask LLM such as ChatGPT, you could sometimes see them pause or gradually create tokens. This is how transformers process different parts of the input data and combine them efficiently to produce results.

4) Halluciaion

⁕ Definition: A phenomenon that outputs a story that is less relevant or made up to a question as an answer.
⁕ Easy but deep explanation: LLM doesn't always have access to the source of all content, it only has access to statistical summaries where that information is reduced. Even though you can refer to multiple materials, there is no guarantee that you will always find reliable information. That's why you can't copy the information like you did Ctrl-C and Ctrl-V. Resolving this flaw is becoming an ongoing research topic.

5) Generative Adversarial Network (GAN)
⁕ Definition: a model commonly used in image Generative AI, consisting of two neural networks.
⁕ Easy but deep explanation: Generative adversarial networks consist of two components: a generator neural network and a discriminator neural network. These two neural networks train against each other to generate more reliable new data. The role of the generator is to create images that are as similar as those in the training data, and the discriminator becomes an evaluator to distinguish between the generated images and those in the training data. When a generator creates a poor image, the discriminator easily recognizes that it is not a good result. The discriminator provides feedback to the generator to ensure that the generator's performance continues to improve.

The second part contains information on how Generative AI will develop in the near future and what areas humans should think about in order to improve their Generative AI day by day. The second part will be released on Thursday, Sept. 12 with the views of Jerry Kaplan, a world-renowned artificial intelligence authority!

Subscribe to our newsletter