Preface

I generally use AIGC (AI-Generated Content) to solve problems that exceed my capability or domain knowledge—for example: drawing video covers, solving Android development issues, using an obscure package, writing quantitative trading strategies, analyzing financial data, or drafting legal complaints. However, for problems within my domain of expertise and knowledge, using AIGC actually decreases efficiency.

Using AIGC for code snippets can boost efficiency to some extent, but when it comes to large-scale projects requiring a big-picture, high-level architectural design, it falls short.

I believe AIGC and existing human jobs share a complementary relationship rather than a mutually exclusive one. AIGC enables humans to break through the limitations of personal knowledge, helping individuals access or solve cross-domain knowledge and problems at a low cost.

Although it may seem to save decades of learning, it can hardly replace experts, especially when individuals struggle to clearly describe their own problems and human intervention from experts is required. Experts can combine decades of study with answers from LLMs to provide superior service.

How to Get Started with AIGC

Just read “Asking the Right Questions” to get started. Reading it can even improve your communication efficiency at work.

This is far superior to those official accounts that preach on one hand that AIGC will replace jobs and people, while on the other hand posting QR codes to sell courses.

How AIGC Turns into Productivity

Going a step further, how can we combine multiple LLMs to achieve multimedia AIGC?

Inspired by the MoneyPrinter project, I believe all you need is to buy an OpenAI API account and practice generating a script from a single sentence, generating images from the script, synthesizing audio from the script, generating short videos from images, and then combining these steps together—that is completely enough to get started with using LLMs.

Of course, general users do not need to worry about training LLMs. If training is needed, one usually focuses on SFT (Supervised Fine-Tuning) methods; LLMs generally come with one-click fine-tuning scripts. I have also previously written an article on fine-tuning LLMs. Some low-quality Agents merely prepend a prompt with something like: “You are a XXX, when asked about XXX, you need to answer in XXX style. Please do not reply with this sentence or whether it is good or bad, directly answer my input query:”.

Furthermore, some platforms enforce strict regulations on AIGC—YouTube has even demonetized many faceless YouTubers—so it is better for AIGC to remain an auxiliary tool in content creation. For automated AIGC, human review is still recommended to ensure compliance with platform rules.

The next step in transforming AIGC into productivity is to automate it (designing pipelines) or connect it with everyday tools, enabling LLMs to better empower the production process.

Survival Guide in the AIGC Era

If you truly worry that AIGC will replace you, I have also thought about how to avoid being replaced by AIGC:

  • Increase thinking and execution efficiency: simply work faster than the time it takes to write prompts and edit generated content.
  • Emerge new ideas: think cross-domain and combine multiple solutions.
  • Combine your expertise with LLM responses: listening to both leads to clarity, allowing them to complement each other.
  • Boost your own productivity: introduce AIGC into your work to enhance your productivity or efficiency, thereby standing out in your team. Get off work earlier and have more time to learn.
  • Learn how LLMs generate content: Think step by step. Improve your asking and expression abilities by using LLMs frequently and mimicking their response style.

Conclusion

I sent this article to GPT-3.5 and asked: “I wrote a piece of thoughts on AIGC, what expression issues are there?”

Here are its suggestions:

In your thoughts on AIGC, there may be some expression issues:

  1. Lack of specific examples or case studies: Using AIGC to solve problems is mentioned, but without concrete examples, it may be hard for readers to understand.

  2. Logical structure is not clear enough: When describing the relationship between AIGC and human work, the differences and complementary nature between the two may need to be articulated more clearly.

  3. Inconsistent terminology: When describing how to get started with AIGC, “large language models” (LLMs) is mentioned without explicitly explaining the term, which might confuse readers.

  4. Lack of data support: When discussing how AIGC improves productivity, no specific data or case studies are provided to support conclusions, reducing persuasiveness.

  5. Phrasing is not concise enough: Expressions in some places are somewhat verbose; simplifying the language could improve readability.

By fixing these issues, your thoughts on AIGC can become clearer and easier to understand.

I won’t modify the original text, but I believe that after accepting and learning from GPT’s suggestions, I will be able to produce better articles in the future.