The AI ​​models are becoming smarter according to the day. Whether it is resolving arguments or equations, LLMS (small for big language models) is proving to be the clever child in the classroom today. But this is coding where things are really getting interesting. Large language models (LLMS) such as GPT-4, Cloud 3.5 Sonnet, and Gemini 2.5 Pro are already writing code and debugging like professional couders. And what started as AI assistants for the programmer is now increasing the concerns of completely changing the human coder. A senior AI researcher from Alibaba believes that this future is not far away.

An employee research scientist of Alibaba’s Quven team was Binuan, recently shared his views on X (East Twitter), arguing that “LLMS would compulsorily cross humans in coding.” Not only is AI Mirroring how humans learn, but it is also getting so advanced that it is also excluding humans.

In his post, Hui says that human coders usually pass through two stages. First, according to him, there is a memoir and imitation. Where coders learn syntax, study examples, and repeat good projects. The second stage is testing and error. In this phase, programmer writes code, runs it, fixes bugs, and improves through feedback.

Now with LLMS being smarter, HUI argues that AI is also following the human path to learning. The pretraying model is allowing the huge amount of code to absorb and compress and compress that is actually far ahead that can probably miss any human. “Strengthening learning (RL) aligns the model with execution through feedback, but humans on a speed and scale cannot match, are completed with millions of rollouts in a short time,” Writes.

He further states that when an individual programmer can debug a project at a time, LLMs are able to run millions of recurrents in parallel, allowing them to accelerate their growth curve. “In parameter capacity and recurrence it means that it is only a matter of time before the models move beyond humans,” they say.

Hui’s prediction does not fall into isolation. A recent study by a Stanford University has already revealed the initial signals of this change in the workforce. Research, which analyzed payroll data from millions of American workers, found that entry-level software engineers have been the most difficult hit since the launch of Chatgpt in late 2022. The study showed a 16 percent decline in employment among workers between the ages of 22 to 25, such as in industries such as coding and customer service.

According to the study, instead of hiring fresh graduates, companies are bending to AI for senior engineers maintaining or even extending the roles for repeated coding works. In practice, AI is working as an assistant to experienced developers, but researchers in Stanford warned that these similar forces could eventually expand to more advanced level work.

But the capacity of the code is just the beginning. Hui says that with continuous improvement and training, the AI ​​models can begin to prove themselves. He suggests that once models reach the stage of Artificial General Intelligence (AGI) – the levels at which companies like Meta, Openi and Google are running – they will not only be able to follow the signs, but will be able to think and work like humans. At that point, they argue, AI will not only write, but will also optimize their own code. “Code is not only the foundation of human productivity, but also the initial point of recurring improvement. When models can write and adapt their own code, what we see is no longer AGI, but the first signs of ASI. AI4AI is in the way.”