The World Artificial Intelligence Conference (WAIC) revealed one trend: large AI models are leading a technological race known as the "battle of a hundred models."
As a company that has always emphasized algorithms in the industry, MeDomino is often asked about AI, AIGC, LLMs, and related fields, so we decided to share some thoughts. To summarize first, our core view is: AI will be everywhere, but ChatGPT will not be everyone's advantage, except OpenAI's.
01 AI Will Be Everywhere
Anyone who has experienced GPT-4 can feel this, not to mention the recently opened Code Interpreter. In the past two years, AI-based applications have appeared one after another. In the intense "battle of a hundred models," some contributed classic jokes, while others teased that the core competitiveness of domestic large models was their naming.
Large models in the WAIC exhibition area. Source: photographer Shen Yuan, all shot onsite at WAIC.
MeDomino's investment in AI began in 2019. Today, we have completed training data construction, model training, fine-tuning, and testing, forming a vertical-domain model for the life sciences industry. Before ChatGPT-3.5 was released at the end of 2022, MeDomino was an early practitioner of AI model applications in the industry. When most industry applications still relied on manual work, MeDomino had already used AI to replace manual work in specific task scenarios.
It was like taking up a bow and arrow first in an age of primitive tools, giving MeDomino a huge advantage.
But once GPT-3.5 appeared, the situation became one where everyone could quickly get a machine gun. It was not very accurate at the time, but the threshold was low, the cost acceptable, and most importantly, it was loud.
What happened next is familiar: GPT-4 greatly surpassed GPT-3.5, and the machine gun became much more accurate. Today, in terms of capability, GPT-4 is clearly far ahead.
The AI technology breakthrough means many companies no longer need to go through the painful process we experienced: data cleaning, annotation, training, iteration, and more. The advantage we paid a great cost to build was quickly flattened. We were shocked and confused.
So can everyone use ChatGPT to form their own advantage?
02 ChatGPT Will Not Be Everyone's Advantage
Let us first think about one question. In the PC era, the Wintel combination dominated the market: Intel CPUs at the core and Microsoft Windows as the operating system. For the many PC products relying on Wintel, where did core differences lie? In most cases, I believe the differences were mainly industrial design and marketing.
Compared with hardware products, software products such as games need an operating system to run, but do not depend on the system for their own value. Product upgrades can be self-sustaining and even mutually reinforcing with the system:
- Game upgrades can better adapt to the system and give users a better gaming experience.
- System upgrades can provide a smoother operating environment for the product.
So, returning to the beginning, why do I say ChatGPT will not be everyone's advantage?
Just as an operating system relates to a PC, many emerging large-model products rely on ChatGPT itself. Their product capability upgrades depend largely on ChatGPT's upgrades. Yet they are not exactly the same as PCs, because ChatGPT upgrades may also lead to their disappearance. After all, none of us knows how many native capabilities the next version of ChatGPT will extend.
This is also what MeDomino worries about. We do not want to be a PC that upgrades only by relying on system upgrades. We are committed to becoming an independent, self-upgrading game.
Some friends may know that our content management platform, MeDomino Content Hub, initially focused on tagging capabilities, including title-level tags, summary-level tags, module-level tags, and self-maintained tag capabilities. It solved the content storage and retrieval problems that troubled many pharma and medtech companies. We did not expect ChatGPT to appear and potentially flatten some of these foundational capabilities to a certain extent. In a sense, ChatGPT even became our competitor.
To be honest, during that period we were anxious. After the anxiety, we decided to connect ChatGPT's capabilities and turn competition into combination.
Now, beyond the most fundamental tagging capabilities, MeDomino Content Hub can automatically deliver advanced capabilities such as paragraph-level summaries, page-level summaries, content interpretation, and modular decomposition. Combined with AI-assisted physician mindset data, it can precisely identify content suitable for physicians, quickly produce and distribute content, and achieve personalized content engagement at scale.
We are even exploring applications such as automatic content generation.
03 Final Thoughts
In MeDomino's game world, iterations in AI technology will only continue to help us optimize product experience. We do not reject any new technology, as long as it helps us better serve the life sciences industry and benefit patients.
We will adopt ChatGPT, but ChatGPT will not become us.
A game runs on a system, but the system does not become the game.
AI will eventually be everywhere, but AI will never become humanity.
