What is Generative AI? ChatGPT, DALL-E, Codex, and Beyond

Published in

Startup Island TAIWAN

6 min readApr 28, 2023

“Baby Artificial Intelligence” Generated by Canva’s Generative AI

ChatGPT: The Evolving AI Landscape

For most of us, a year ago, the most commonly heard AI tools were personal assistants such as Siri, Alexa, and Google Assistance. Since its release on November 30th, 2022, the generative AI named ChatGPT has gained widespread attention in a flash as a state-of-the-art language model. However, many may not know that ChatGPT had three iterations before itself; ChatGPT is actually version 3.5, named GPT-3.5. The journey started in 2018, with the release of GPT-1 by OpenAI, followed by GPT-2 in 2019, then GPT-3 in June 2020. Just two months after ChatGPT’s release, it achieved the notable milestone of 100 million users. This astonishing growth rate surpassed popular social media platforms by multiples. TikTok took nine months to reach this milestone, and Instagram took 30 months. ChatGPT’s unique pace of scaling has likely made it the fastest-growing web application since the inception of the World Wide Web. According to UBS analysts, “In twenty years following the Internet space, we cannot recall a faster ramp in consumer Internet app.”

The outstanding success case of ChatGPT has triggered a race within the field of AI, with tech giants and countries vying to create their versions. OpenAI released a new iteration of ChatGPT, Microsoft, and Google have announced breakthroughs and begun testing similar products, and Taiwan sets to launch its version by the end of 2023. The rapid adoption and attention the world has seen with ChatGPT have undeniably reshaped the modern landscape of AI technology, especially within the space of Generative Artificial Intelligence (GenAI).

What is Generative AI?

Alan Turning laid down the fundamental groundwork of machine learning (ML) in the field of computer science as early as the mid-1900s with his published paper “Computing Machinery and Intelligence”. Generative AI (GenAI) is an AI system that generates new outputs in response to prompts. The outputs generated derives from the datasets it has been trained on. In contrast to traditional AI systems, designed to identify patterns to create analytics and forecasts, generative AI produces brand-new content creation in the form of text, images, audio, code, and more. This is achieved through a deep learning framework named generative adversarial networks (GANs) that consists of two types of neural networks, a generator and a discriminator.

The generator’s function is to produce new data, whereas the discriminator is to evaluate the data produced by the generator. These two neural networks work together in a loop-like structure, with the generator producing content and then passing the generated content to the discriminator for evaluation. Based on the discriminator’s feedback, the generator improves the content, and the process repeats itself until the content generated is as close to actual data as possible.

Let’s see how ChatGPT explains generative AI:

“Generative AI, powered by complex algorithms and deep learning techniques, is a cutting-edge technology that can create original content without explicit programming. It includes techniques like Generative Adversarial Networks (GANs) and has the potential to revolutionize industries such as art, design, marketing, and advertising. However, ethical considerations such as responsible use, copyright infringement, and bias in data must be addressed. Despite the challenges, generative AI is poised to shape the future of technology and creativity, offering exciting possibilities for innovation and growth in various industries.”

Taiwan’s Stance on Generative AI

The global race for AI has been in full swing ever since the initial release of ChatGPT. On top of GPT-4’s release by OpenAI, tech giants like Google and Microsoft have progressed significantly with integrating generative AI into their applications. Wu Tsung-Tsong, the Minister of Taiwan’s National Science and Technology Council, stated that Taiwan is set to release its own ChatGPT by the end of 2023. However, rather than creating a general system like ChatGPT, Minister Wu explains, differing from its predecessor, Taiwan’s version would not aim to create a general system, but instead, it would target specific tasks.

Sega Cheng, Co-Founder & Chairman & CEO of one of Taiwan’s largest AI solution startups, iKala, agrees with Minister Wu’s approach. He expressed that creating a similar system to ChatGPT would be a waste of resources; instead, Taiwan should develop related applications. Mr. Cheng also explained that the development of AI requires in-depth practical skills and expertise; the government should act as a facilitator connecting stakeholders to explore future possibilities as opposed to making unilateral decisions.

Ethan Tu, the founder of Taiwan AI Labs, shared a similar perspective. During a separate interview, Mr. Tu suggested that Taiwan indeed has the ability to create and train an AI language model; however, instead of creating something like ChatGPT, the creation of Taiwan should be catered to the country’s specific needs for it to be meaningful. To assist this process, Mr. Tu said a coordinator would be necessary to bring together local language databases for training.

One of the constraints Taiwan faces in the development of its generative AI is cost. An estimated investment of US$ 10 billion was invested in developing ChatGPT, while Taiwan’s annual technology development budget is limited to US$ 4.34 billion. However, as AI technology continues to evolve, the required development costs should reduce correspondingly. An excellent example is the introduction of Standford’s Alpaca 7B model, a model fine-tuned from Meta’s LLaMA 7B model. Although it was created at a fraction of the cost (less than US$ 600), Stanford researchers said it behaves “qualitatively similarly” to ChatGPT.

The exponential developing pace of AI development is evident. Since ChatGPT’s release to the public in November 2023, we are already being flooded with AI tools with various capabilities. Forecasters from Our World in Data predict by 2030, there’s a 25% chance of the first artificial general intelligence (AGI) being developed, tested, and publicly announced, and a 50% chance by 2050. Last month, in March 2023, a research paper published by Microsoft’s researchers suggested that GPT-4 displays early signs of AGI. This means that GPT-4 shows capabilities equal to or beyond human intelligence. With the exponential growth of AI technologies, the question arises, how soon will Taiwan catch up and release its version of generative AI, and at what cost?

Below are some generative AI tools to explore right now:

GPT-4

GPT-4 is the successor of ChatGPT, which is also a large language model (LLM) developed by OpenAI. GPT is an abbreviation for “generative pretrained transformer”. Released in March 2023, GPT-4 can solve complex problems with higher accuracy. In addition, GPT-4 now also accepts image inputs, longer text content inputs and can produce more creative content.

DALL-E 2

DALL-E 2 is a generative AI that produces images from text prompts. It can generate original images and art, expand beyond already existing images, edit existing pieces by adding or removing components, and create various variations of images from an original piece.

Codex

Derived from GPT-3, Codex is a collaboration between OpenAI and Github. The system can understand natural language prompts, recognize problem statements and intents, then generate corresponding codes in various programming languages. It has been trained in Python, Javascript, PHP, Shell, Perl, and others.

Dadabots

Dadabots is a generative AI-powered online platform created by musicians interested in science and machine learning, CJ Carr and Zack Zukowski. The platform can generate music pieces that imitate different genres and styles of music by famous bands such as Nirvana.

—

References

AI Labs’ Tu calls for development of AI unlike ChatGPT

https://www.taipeitimes.com/News/biz/archives/2023/04/03/2003797201

AI timelines: What do experts in artificial intelligence expect for the future?

https://ourworldindata.org/ai-timelines

Alpaca: A Strong, Replicable Instruction-Following Model

https://crfm.stanford.edu/2023/03/13/alpaca.html

ChatGPT may be the fastest-growing consumer app in internet history, reaching 100 million users in just over 2 months, UBS report says

https://www.businessinsider.com/chatgpt-may-be-fastest-growing-app-in-history-ubs-study-2023-2

ChatGPT sets record for fastest-growing user base — analyst note

https://www.reuters.com/technology/chatgpt-sets-record-fastest-growing-user-base-analyst-note-2023-02-01/