A Reflection on General AI (feat. Modern Times)

Experiencing generative artificial intelligence (General AI), the thought that comes to mind is — it feels strikingly similar to the impact factories, computers, and the internet once delivered to humanity. Before making simplistic positive or negative judgments, the practical reality is that some people will lose their jobs while others will gain new ones, and what is even clearer is that each successive stage will widen the gaps between people. 

In 1913, when the conveyor belt entered the world, the division and specialization of labor of that era ignited human productivity. But then, the standardization required for mass production inevitably came into conflict with the worker as a 'human being' who dislikes uniformity. Stated a little dramatically, humans were reduced to assistants supporting machines, and in this process, the individuality and diversity of workers were routinely ignored.     
What we commonly call the Industrial Revolution and the Information Revolution — I find myself thinking these are not really about machines that labor in place of humans, computers that calculate in place of humans, the internet that delivers and manages information in place of humans, or AI that thinks in place of humans. Rather, like the metaphors in Charlie Chaplin's Modern Times, each of those bends in the road feels like its own new ideology. 

As Yuval Harari describes in Sapiens, humanity survived through the cognitive revolution and the linguistic revolution, and through competition among the survivors, we continue to create new languages. And the agents driving this change are shifting their place and scope — from individual to collective, from natural person to legal person, from offline to digital. This resembles a kind of matrix-like structure, or perhaps compounding interest, or maybe synapses — connecting and expanding exponentially. 

The field of generative AI took on a popular form last year starting with DALL·E 2 led by Elon Musk?, followed by the release of Stable Diffusion, Midjourney and others. Starting from this realm of generating images from text, the General AI domain that had been stirring has been expanding wildly? since the release of ChatGPT last month. The biggest difference from the earlier generative models mentioned above is that ChatGPT offers a universal interface (conversational form), and through this, there is an explosive expansion of third-party applications, and what deserves even more attention here is that, unlike the App Store ecosystem of the past, many of these third-party applications are being built and deployed using no-code.

In a previous post, the reason I described it as the crisis of the developer rather than the programmer was precisely this point. Rather than which language and platform you can implement, or whether you are front-end or back-end — more fundamentally, I want to emphasize that the more important capability is whether you can implement what you genuinely think is needed, by yourself. I hope readers won't be confined to familiar programming languages or to the type of output (app or web). Even with years of IT experience, in many cases people can't even dream of building their own service. Most have only an isolated part — front-end or back-end, or planning, or design. Those rare individuals who can compose a service single-handedly are sometimes called full-stack. But most of those projects are large-scale B2B work. The cases where someone actually builds a micro-service that they themselves or those around them can use, and then deploys it for the people around them, are extremely rare. I suspect this is the danger that comes from standards, frames, and accumulated experience. I hope this becomes an opportunity to reconsider once again the basic? specifications and conditions implied when we say the word "service." 
 
As I'll share later, most of those developing and deploying recently released applications are not developers. They are simply manufacturing workers interested in AI, marketers, healthcare workers, or undergraduates and graduate students researching AI. Meanwhile, the majority of people in the industry are buried in their day jobs — never getting around to experiencing it, or sometimes repeating their own particular habit of judgment like "that's still inaccurate—," while a new ideology is taking root. And once this ideology settles in, perhaps we'll resent the world while yielding our place, or belatedly find ourselves repeating a life with no afternoons. With that very personal sense of alarm, I'm sharing the references I tested and organized.

Reference 

For reference, after testing so far, I've come to think that the most important point in dealing with generative AI is how you use the prompt and how much you can reuse the content of those prompts and their results as your own database — that's the key.
The prompt here is structured around the idea that, just as questions matter in conversations between people, the intent and manner of questions also matter in conversations between AI and humans. And to add one more remark — just as in a conversation with someone, the reason you feel frustrated or fail to get the answer you want is sometimes a problem with you and not with the other party, in conversations with AI as well, I hope readers won't fall into the habit of judgment that simply blames the AI's incompleteness, the physical constraints of its database, or the limits of its inference. This habit of judgment most often arises when one is positioned at the center of an organization, and when that happens, a small individual perception can unintentionally become a case of horichilli 毫釐千里 (a thousand-mile divergence from a hairsbreadth start), and thereby steer the fate of the entire organization. 

Prompt builder and management
https://riku.ai/

A service that creates multi-page slides from a single sentence
https://beta.tome.app/

A tweet proposing a guide for prompt construction 
https://twitter.com/thatroblennon/status/1615104249192488980?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed&ref_url=https%3A%2F%2Fwww.notion.so%2Fthinknormal%2FGeneral-AI-ChatGPT-etc-c4519c685846410bb9666575ac819bf3 

A post on writing prompts that account for SEO (search engine optimization) 
https://linkshero.com/chatgpt-for-seo/

Research paper on writing prompts to handle large documents exceeding 3,000 characters
https://arxiv.org/abs/2110.01691

Cases of using ChatGPT

A service that creates fairy tales for lullabies?
https://twitter.com/LinusEkenstam/status/1615715273432080388?s=20&t=3lrPrhHDTzarf-ai7NqZuw

Ammaar Reshi published a children's book in 72 hours (ChatGPT + Midjourney)

AI podcast

A case of using ChatGPT as a guide for writing a startup support program application

A case of summarizing an academic paper

I'll close this post by quoting from Han Byung-Chul's The Disappearance of Things, a book I've recently read with deep impression.

Artificial intelligence learns from the past. The future that AI predicts is not a future in the true sense. AI is event-blind (ereignisblind). Thinking, by contrast, has the character of an event. Thinking places something entirely different into the world. What AI lacks is precisely the negativity of rupture that allows something genuinely new to begin. AI ultimately continues sameness. --- p.67