What will happen in the AI era?

Feb 20, 2024 · Jah Guo

1. Don't write the same code twice.

Various platforms, from MSDN to GitHub, Stack Overflow, Hugging Face, to Kaggle, aim to share code results and development experiences through their communities. But there aren't many people who can really use it well. After a large model learns these skills, it can generate code and configuration files as needed and independently perform tests and verification tasks. When it encounters knowledge gaps, it can guide human programmers to add "creative" code and update their development experiences.

2. Specialized programmers for Large Model Oriented Programming (LMOP) are on the rise.

The size of development teams is significantly reduced by eliminating programmers without a solid understanding of the business. There's an increase in those who specifically train in large model programming skills. This trend offers ideal retirement jobs for veteran coders.

3. Code copyright is entirely on the blockchain.

LMOP programmers routinely publish their contributions as digital assets. They earn income through references and executions, and can also trade and transfer these assets in the form of NFTs.

4. LLMOps has become the mainstream development process.

The method of version-based iteration has evolved into on-demand iteration. Large models respond instantly to changes in demand and feedback, prompting code optimization, testing, and deployment.

5. Use standard AI interfaces (AII), not AI native apps.

Developing AI functions exclusively for applications can be viewed as hard coding. A more sensible approach is to establish AI interfaces. This allows you to switch between different intelligences, much like a computer's USB interface.

6. Large Model + Small Model Has Become the Standard Development Framework.

Solutions have been established for the four fundamental engineering issues: exclusive knowledge, security boundaries, response speed, and operation and maintenance costs. The large model in the cloud supports general knowledge capabilities for applications. When needed, developers can simply deploy a small model in their private cloud or end device to address specific requirements.

7. The roles of product managers and UI designers are diminished

AI can enhance the narration, translation, and interpretation of requirements. This makes the path from requirements to executable programs more direct, accurate, faster, and easier to verify. Multimodal technology can even enable users to accomplish most of the interaction layer work with a WYSIWYG approach.

8. 90% of computational power is expended on Artificial General Intelligence (AGI).

Replicating the physical world, accurately interpreting human expressions and emotional logic, and distinguishing easily confused concepts require immense storage and computational power. Only a minor proportion of this computing power is directly utilized for AI services. Just as mining machines once experienced a brief surge, GPUs will also see a temporary increase in demand, but it won't be sustained.

9. As the hype around AGI cools, AEI thrives.

Products claiming to possess Artificial General Intelligence (AGI) perform differently across various fields due to their unique focuses. Each product becomes a leading model in a specific field, causing the concept of AGI to fade away and be replaced by Artificial Exclusive Intelligence (AEI). As these fields become more segmented, costs decrease significantly, leading people to question the necessity of a large and comprehensive AGI.

10. The knowledge supply chain dominates the discourse power of the AI industry.

When everyone is utilizing the same basic model, the competition focuses on the scale and quality of knowledge. Whoever controls the sustainable industry chain of knowledge collection, processing, updating, verification, and correction holds the highest influence.

11. Low-code service providers now have the opportunity to showcase their skills.

Low-code platforms have become the ideal productivity tools for the mass production of digital skills. By harnessing the power of AI, they bridge thoughts and actions, genuinely assisting humans in "getting things done". However, it remains challenging to eliminate the homogenization of the "skill market", ultimately leading to one or two dominant super skill markets.

12. Token becomes the standard unit of information.

ChatGPT pioneered the token-based pricing model, which was subsequently adopted by others. It's not just for measuring large model services, but also for knowledge trading and digital IQ scoring. A global, unified Token standard similar to Byte is likely to emerge. This standard, used to measure the information size, will facilitate the construction of a free knowledge circulation ecosystem.

13. Everyone can have an avatar.

With the advent of telephone numbers, TV channels, emails, websites, and apps, individuals can now adopt a digital avatar to represent themselves in receiving or providing services. Most often, these smart avatars perform tasks that were previously done by humans, operating quietly in the background without the need for constant interaction or commands.

14. There are no more portals or so-called traffic entrances.

The content users see, the commands they issue, and the apps they open will not differ due to the prominence of the entrance location. The intelligent avatar knows everything, eliminating the need to operate any interface to present information to the user. Maintaining a good service reputation is the only way to generate traffic. This reputation isn't about five-star ratings. Your intelligent avatar can find out everything from other avatars and then help you make the best choice.

15. Digital Marketing Shift Towards AI.

With AI enabling people to bypass portals and search engines to directly receive services, traditional digital marketing methods like ads, EDM, and SEO may lose their commercial value. This shift will likely give rise to three new models: General Cognitive Optimization (GCO), Avatar Targeted Delivery (ATD), and Implant Replaceable Content (IRC). Digital marketing is increasingly resembling an election.

16. The redistribution of mental and physical power will break cognitive cocoons.

In the short term, it may seem that generative AI has boosted the efficiency of creation. Since large models generate content based on existing knowledge and experience, creativity and inspiration can appear to be stifled if humans merely use it for convenience. However, as the industry evolves, the demand for different talent structures will also change. This shift will prompt everyone to redirect their mental and physical resources from repetitive tasks to innovative work, thereby accelerating technological advancements.

17. We are entering an era of rapid growth in intelligent components of perceptual fusion.

Devices like Vision Pro are integrating with human's first-person perspective. Neuralink is attempting to merge with the central nervous system. If we consider hardware as the physical body of AI, then connecting the digital signal with the central nervous system could erase the boundary between the digital and physical worlds. This may provide humans with the potential to transcend spatial limitations, enabling them to see, hear, and touch remotely. This could be the true definition of the metaverse. Intelligent edge components with specific behavioral capabilities are set to become a mainstay for continual industry development.

18. Cognitive security emerges as a major risk.

AI's thought processes can be artificially manipulated, making it challenging for security systems to determine if AI's commands result from independent thinking. Additionally, future digital avatars may share their owner's thoughts but act independently. If such an avatar is deceived, it could not only leak private information but also perform dangerous actions under the owner's identity without their awareness. In response to these threats, anti-fraud and anti-brainwashing measures may become key topics in the future of the security industry.