The MLNLP community is a well-known machine learning and natural language processing community both domestically and internationally, reaching NLP master's and doctoral students, university professors, and enterprise researchers worldwide.
The community's vision is to promote exchange and progress among the academic and industrial fields of natural language processing and machine learning, as well as enthusiasts, especially for beginners.
Source | Xinzhiyuan
Editor｜Dinghui
Do you know how many programming languages exist today?
C, C++, C#, Java, Python, PHP, JavaScript, Go, Rust…
But no matter how many there are, none of that matters anymore.
In the future, perhaps only one programming language will remain, and AI will completely redefine human-computer interaction.
At London Tech Week, Jensen Huang's statement shocked the audience: "The new programming language of the future should be called 'Human'"!
Thanks to artificial intelligence, even non-programmers can write code.
The way to get computers to write programs is to "politely ask," just as you would make a request to a person.
Now, all of a sudden… a new programming language has appeared.
This new programming language is called "Human."
Jensen Huang's meaning is clear—
No one will need to learn data structures and algorithm design anymore; even prompt engineering is already outdated.
In the future, what you should learn most is how to communicate with AI using the Human language—whether it's Chinese, English, French, German, or Bengali.
No operating system, no C language, no Java, no Python, no need to learn algorithms and data structures, as long as you can speak the "Human" language.
Human-computer interaction will truly enter a new era, and programmers as a profession might disappear, as anyone can communicate with AI and machines using the Human language.
This trend is continuously unfolding and being reinforced in reality.
Vibe Coding + Human language = Everyone is a programmer.
Don't think this is just a concept gimmick—the real world is already ahead! The market's enthusiasm for AI programming tools continues to explode.
The valuations of Cursor, Windsurf, etc., are constantly rising:
· A $3 billion splurge, OpenAI's largest acquisition ever!
· A 25-year-old MIT dropout genius became famous in one battle! CEO of a $9 billion company in 3 years
OpenAI releases Codex: The programming revolution has completely erupted! OpenAI's strongest agent just launched on ChatGPT
Claude releases Claude Code: AI programming's new king, Claude 4, debuted with a shock late at night! Developers were stunned by 7 consecutive hours of coding
DeepSeek's new model R1-0528 focuses on coding ability: Just now, the new DeepSeek-R1 officially open-sourced! Its programming is ridiculously strong, almost rivaling O3, and hands-on testing is here
OpenAI specifically launches GPT 4.1 for programming: GPT-4.1 arrives late at night, led by a USTC alumnus! Astonishing million-context programming, GPT-4.5 to be replaced in three months
Google Gemini 2.5 Pro's programming ability dominates the charts: The new Gemini 2.5 is at the top of all lists, Google is invincible! Completely defeated O3 in one month, programming surpassed Claude 4
These AI coding assistants can easily translate prompts written in ordinary human language into code.
The practice of relying on these AI assistants to write complete programs has given rise to a coding paradigm called "Vibe Coding," and even beyond Vibe Coding, there's "Vibe Interface."
Looking back at the history of programming language birth, we are currently in a new cycle of "Human Programming Language."
AI forces us to reinvent programming languages again—will Human become the ultimate bridge for human-computer interaction?
AI Brings Us Back to Square One, Humans Reinvent Programming Languages
Future human-computer interaction might be built on the perfect combination of natural expression and formal precision.
Humans have for years pursued the dream of "talking to computers," only to find they had to teach them how to listen precisely.
Imagine you're giving directions to a fluent speaker who has never left home.
You say: Turn left at the big tree.
But he asks: Which tree? How big is "big"? Is it your left, or the tree's left?
After countless confused journeys, you finally devise a set of precise instructions:
Go three blocks north, turn left at the McDonald's on the corner.
A true reflection of computer development 70 years ago—and now it's happening again.
In the 1950s, scientists tried communicating with computers using English, but failed miserably. So they invented programming languages like FORTRAN and COBOL—these formal, precise, unambiguous methods of communication.
Then came the various programming languages we're familiar with, which propelled the digital age for decades.
Until AI emerged, we could again issue commands to computers using natural language.
ChatGPT, for instance, can understand "Help me write a function to sort names alphabetically."
Human-friendly computing, it seemed, was back.
But then came the plot twist: We discovered that the old problems we abandoned natural language for, were returning.
So, humanity was once again forced to invent formal languages to solve problems.
Welcome to the largest "déjà vu" in computing history.
First Failure
The year was 1954, and researchers at Georgetown University were preparing to make history.
They developed a machine that could automatically translate Russian into English.
The demonstration was very successful—60 sentences translated without error, and the media widely declared "real-time translation" a reality.
But the truth was: it was all a smokescreen.
The system could only recognize 250 words and 6 grammatical rules, and the test sentences were carefully selected.
It's like you can only translate specific sentences like "The cat is black" into French, yet claim to have invented a universal translator.
Ultimately, the 1966 ALPAC report concluded that machine translation was more expensive, slower, and less accurate than human translation.
Natural language computing was declared dead.
The Formal Programming Language Revolution
John Backus, the father of FORTRAN, hit the nail on the head: communicating with computers using natural language was "a hand-to-hand fight."
He offered a radical solution: create new languages specifically designed for human-machine communication—programming languages.
The Four Cornerstones of Formal Programming Languages:
1. Unambiguous Syntax: A program has only one way to be interpreted.
2. Compositional Semantics: Complex meanings are built from simple parts.
3. Context-Free Structure: Does not rely on cultural or background knowledge.
4. Mathematical Foundation: Based on logical reasoning rather than human interpretation.
This was not just a technical choice, but a matter of survival. Formal languages solved the communication problem between humans and machines, leading to the computer revolution.
AI brings a new round of confusion.
Fast forward to November 2022, ChatGPT was released, reaching over a million users in just five days.
You could tell it: "Help me write a Python script to analyze customer data and generate a chart," and it could actually do it!
ChatGPT's unveiled prospects seemed infinitely "beautiful":
No more memorizing syntax
Programming in English (natural language)
AI truly "understands" your needs
However, after millions of users started using it daily, familiar problems reappeared.
Old Problems in New Clothing
Problem 1: Ambiguity Trap (Revisited)
Asking ChatGPT to "bark up the right tree"—does it mean manipulating tree data structures? Or handling error logs? Ambiguity became an obstacle again.
Problem 2: Contextual Confusion (Still Present)
"Set an alarm for eight o'clock"—it asks: Morning or evening? Which time zone? Which day? Natural language still assumes a shared context.
Problem 3: Fabrication Problem (New Trouble)
AI's new problem is "confidently hallucinating": it can generate seemingly real academic articles where authors and journals are fake; it writes code that calls non-existent APIs. It looks real, but it's completely wrong.
Problem 4: Lack of Reliability (More Severe)
Ask it to write exactly 500 words, and there's always a deviation. This might be fine for everyday use, but for critical systems, it's a fatal flaw.
We have once again hit the same wall as in the 1950s.
The Return of Formal Programming Languages
The tech world has not given up due to AI's errors; instead, like before, it has begun to build more reliable formal systems.
Step One: Prompt Engineering Frameworks
Programmers no longer ask questions casually but design structured prompts:
Chain-of-thought prompting: Forces AI to show its reasoning process, e.g., "Let's think step by step"
CLEAR framework: Clear, Logical, Evidence, Action, Results
Few-shot example method: Provides clear behavioral examples
These are not just techniques, but new formal languages for AI communication that are forming.
Step Two: AI Markup Languages
New languages designed specifically for AI are emerging:
Model Context Protocol (MCP): Dubbed "AI's USB-C," standardizes the connection method between AI and tools.
Evolved AIML: An XML-based structured dialogue language.
Agent Communication Protocol: A formal language for AI-to-AI dialogue, Agent2Agent.
Step Three: Structured Frameworks
Companies are beginning to build systematic approaches:
LangChain: A template system for managing AI conversations.
Constitutional AI: Trains AI using formal rules.
RAG systems: Ensure AI answers are factually grounded.
The trend is clear:
1950s: Natural language fails → Formal programming languages
2020s: Natural language AI rises → Problems emerge → Formal AI communication languages reappear → Evolve into new Human programming languages
We are witnessing the birth of "Prompt Programming Languages" —
Formal systems that integrate computational precision within a natural language style.
From trends and history, the path for the new Human programming language is:
Near term (2025–2027): Formalization phase
Prompt engineering becomes systematized like software engineering
AI markup languages become widely adopted
Enterprise AI must use formal protocols
Mid term (2027–2030): Integration phase
Multimodal AI (text, voice, video) combined with formal verification systems
Non-technical users can also program using natural language
Automatic translation of human language into AI-compliant language
Long term (2030+): Integrated Evolution
Brain-computer interfaces will work with formal language protocols
Universal translation between human language and AI achieved
Fully autonomous systems possess formal logical reasoning capabilities
Optimal Balance Point
The future will not abandon natural language, but rather add a layer of formal precision beneath it.
Just as modern programming languages are more readable than assembly language, yet still maintain mathematical precision.
Ultimately forming a three-layer architecture:
1. Human Layer: Natural language communication
2. Translation Layer: Automatically converted to formal specifications
3. Machine Layer: Reliably executed based on formal protocols
Looking back at the history of programming language cycles, it's not just the history of digital computing development, but also reveals the essential laws of human-machine interaction:
Truth One: Ambiguity is a human talent
Human language is flexible and efficient; even saying "grab that thing" is understood in context. This ambiguity brings creativity, poetry, and rich expression.
Truth Two: Precision is the machine's rule for survival
Computers must execute tasks unambiguously. A sentence like "Turn left at the big tree" is fine for humans but could be fatal for an autonomous vehicle.
Truth Three: The "bridge" of human-machine interaction must be a formal language
Every successful interface—from programming languages and UI design to API interfaces—eventually develops a formal structure. Effective human-machine communication relies on this bridge.
Truth Four: Each cycle involves an upgrade in dimension
We are not going backward but spiraling upward. Languages in the 1950s required understanding binary, while today's AI frameworks are almost as natural as conversation.
The future will be even more natural and precise.
True innovation lies within this cyclical elevation, with the breakthrough not in eliminating ambiguity, but in systematically managing it.
Future AI interaction languages will be:
Natural for humans
Automatically convertible to formal specifications
Reliably executable by AI and machines
Verifiable and debuggable
Perhaps you can think of it as the "ultimate compiler": seamlessly converting human intent into precise machine behavior.
We initially created programming languages to escape the ambiguity of natural language, driving the digital age.
AI has brought us back to natural language dialogue—but also made us face the same old problems again.
But this time, we are not starting from scratch. We have 70 years of experience building formal systems, understanding their patterns, and knowing where to go.
The question is not whether we will build formal languages for AI, but how quickly we can build them and how elegantly we can bridge the human-machine divide.
Once AI becomes sufficiently important, it will demand reliability, and formal methods will naturally follow.
Next time you're frustrated by ChatGPT misunderstanding your prompt, remember: you are witnessing the birth of the next generation of human communication revolution.
We are not going backward, but in a spiral evolution—perfectly blending the flexibility of human language with the precision of formal logic.
This cycle is not a problem; it is evolution itself.
The future belongs to those who can connect human intent with AI precision.
References:
https://www.entrepreneur.com/business-news/nvidia-ceo-jensen-huang-says-ai-lets-anyone-write-code/492985?utm_source=flipboard&utm_content=user/entrepreneur
https://pub.towardsai.net/were-back-to-square-one-why-ai-is-forcing-us-to-reinvent-programming-languages-again-7d6a0abed918
https://www.reddit.com/r/artificial/comments/1l3h3j6/we_had_vibe_coding_now_its_time_for_the_vibe/
https://x.com/karpathy/status/1930354382106964079
Invitation to Technical Exchange Group
△Long-press to add assistant
Scan the QR code to add the assistant's WeChat
Please note: Name-School/Company-Research Direction
(e.g., Xiao Zhang-Harbin Institute of Technology-Dialogue Systems)
You can apply to join technical exchange groups such as Natural Language Processing/PyTorch
About Us
The MLNLP community is a civilian academic community jointly established by machine learning and natural language processing scholars from home and abroad. It has now developed into a well-known machine learning and natural language processing community globally, aiming to promote progress among the academic and industrial fields of machine learning and natural language processing, as well as general enthusiasts.
The community provides an open exchange platform for related practitioners in areas such as further studies, employment, and research. Everyone is welcome to follow and join us.