AI Can Do That Now? A Reality Check on Today’s Capabilities

1. Some Catalysts

Lately, I’ve been interacting a lot with various small and medium-sized companies, and I’ve noticed that all the tech-savvy bosses are incredibly focused on AI. They’re all exploring how to apply AI technology within their companies.

One e-commerce company had a particularly insightful idea. They want to use AI to create a fully intelligent e-commerce operation chain. Starting with product selection using AI and big data, then generating product copy and description images with GPT and ComfyUI, and finally, using RPA for automated monitoring and management of sales data.

So, I decided to organize my thoughts on the current state of AI applications in various fields.

2. AI No-Code Programming: A “Potential Star” Still Learning to Walk (Maturity: 20%, Currently Just a Toy)

Occasionally, I see social media bloggers posting videos saying they’ve tried a certain AI tool, created a very standard and beautiful webpage, or developed a complete app for you. Then they start hyping things like, “Ordinary people can be programmers too, technology isn’t an issue, etc.”

Every time I see these kinds of videos, I find them quite absurd. The so-called webpages they claim to have made are just simple combinations of HTML, CSS, and JavaScript; the same goes for the apps.

Of course, I understand the significant importance of AI in programming, but the prerequisite is that you yourself have a certain technical foundation and can identify errors in the code. When the AI goes off track, you need to be able to pull it back. Otherwise, sometimes the AI might get stuck in an infinite loop of errors, and then you’ll end up cursing AI as rubbish.

3. AI Empowering Programmers (e.g., Cursor): An Efficient “Co-Pilot” (Maturity: 80%)

In contrast to the nascent stage of no-code programming, AI has shown up to 80% maturity in empowering professional programmers. AI programming assistants like Cursor and GitHub Copilot are becoming the “intelligent co-pilots” for more and more developers. They can automatically generate code snippets based on comments or context, complete code, explain complex logic, and even assist with debugging. This greatly improves development efficiency, freeing programmers from tedious repetitive tasks to focus more on architectural design and innovative work. Of course, 80% maturity also means it’s not perfect. AI-generated code can sometimes have potential bugs, security vulnerabilities, or may not be the optimal solution, still requiring strict review and gatekeeping by experienced programmers. It’s more like a highly capable assistant than a complete replacement for human programmers.

4. Digital Humans: Evolving from “Looking Alike” to “Feeling Alike” (Maturity: 60%)

Digital humans, AI-driven virtual avatars, are increasingly appearing in news broadcasting, brand endorsements, virtual customer service, and other scenarios. Digital humans have made considerable progress in “looking alike,” with continuous improvements in the realism of their appearance, expressions, and movements. However, there’s still significant room for improvement in “feeling alike”—that is, achieving natural and smooth interaction, displaying genuine emotions, and deep thinking capabilities. Current digital humans might be able to fluently deliver pre-set scripts or answer common questions based on a knowledge base, but when faced with open-ended conversations, complex emotional exchanges, or situations requiring immediate creative responses, their performance often appears somewhat stiff, and people can easily tell the difference from a real person.

5. AI Image Generation: Visual Magicians of Imagination (Maturity: 80%)

AI image generation technology has experienced explosive growth in the past few years, reaching 80% maturity. Tools like Midjourney, DALL-E, and Stable Diffusion can quickly generate diverse and detailed images based on user-input text descriptions.

This is actually a bit hard to evaluate, as it often depends on the model you’re using. Personally, I feel it’s around 80%-90% mature. Currently, there are two main issues: first, human fingers are still prone to errors. Second, while details like skin and facial expressions are already very realistic, there’s still a noticeable difference compared to photos taken directly with a camera.

6. AI Video Generation: New Narratives in Dynamic Imagery (Maturity: 50%)

Following image generation, AI video generation is becoming the new focus, currently at 50% maturity. Models like Sora, Pika, Runway, and domestic ones like “Keling” and “Vidu” can already generate short video clips ranging from a few seconds to tens of seconds based on text or images. This undoubtedly opens new doors for content creation, advertising, film previews, and other fields. However, compared to static images, video generation faces greater technical challenges in maintaining temporal coherence, the physical realism of object movement, logical consistency in complex scenes, and long video generation. Currently generated video clips sometimes exhibit illogical dynamics, abrupt changes in object morphology, or awkward scene transitions.

7. LLM Functions (Assisting Decision-Making/Learning/Writing, e.g., ChatGPT, DeepSeek): The “Super Brain” for Knowledge Work (Maturity: 90%)

Large Language Models (LLMs) have shown astonishing 90% maturity in applications like assisting decision-making, learning, and writing, making them the most dazzling stars in the current AI field. LLMs represented by the GPT series, Claude, and domestic ones like DeepSeek and Wenxin Yiyan are reshaping the paradigm of knowledge work with their powerful natural language understanding, generation, and reasoning capabilities. They can help us quickly draft emails, reports, and code, summarize lengthy documents, provide learning assistance, and even aid in complex decision analysis in specific fields. Their fluent conversational abilities and broad knowledge coverage make them powerful productivity tools.

However, problems still exist. Sometimes, large models can experience “hallucinations,” giving you an article that “looks perfectly fine in terms of format, symbols, wording, etc., but is actually completely illogical and factually incorrect.”

8. AI for PPT Creation: An Intelligent Assistant for Presentations (Maturity: 60%)

Creating PPTs is a daily task for many professionals, and AI has begun to make its mark in this area, with a maturity of 60%. Tools like WPS AI, AIPPT, Gamma, and Tome can automatically generate a preliminary framework for a PPT, recommend design templates, and even fill in some content and images based on the user-input theme or outline. This, to some extent, alleviates the burden of creating PPTs from scratch and improves efficiency. However, current AI PPT tools often struggle to fully meet professional demands in terms of content depth, logical rigor, and design personalization and aesthetic height. They are more like a starting point; the initial drafts they generate still require users to make significant modifications, refinements, and personalized adjustments to ultimately become high-quality presentations.

9. AI Customer Service: The Tireless “Front Desk” (Maturity: 70%, but AI Lacks Physical Authority)

AI customer service is one of the relatively mature areas of AI commercial application, with a maturity of 70%. Many enterprises have adopted intelligent customer service robots to handle common customer inquiries, answer questions, and provide guidance. They can be online 7×24, effectively reducing labor costs and improving service efficiency. However, the parenthetical note “but AI lacks physical authority” points out one of its core limitations: AI customer service cannot handle complex issues that require physical intervention or fall outside its knowledge base. When users encounter difficult problems or need emotional support, the performance of AI customer service is often unsatisfactory and can easily lead to user dissatisfaction. Therefore, the current better model is a collaboration of “AI + Human,” where AI handles standardized, repetitive inquiries, and complex issues are transferred to human experts to achieve a balance between efficiency and experience.

These days, I’ve seen many companies building local knowledge bases and exploring various possibilities for AI, such as using AI to replace human customer service, or to downsize or reduce the scale of human customer service teams.

But there’s a problem we also encounter when we look for customer service online: people always hope to be able to contact a real person, either at the beginning or eventually.

For example, when there’s an issue with a product or service you purchased on an online platform and you’re seeking a refund or compensation, only a “real person” can handle such a situation.