Alibaba research scientist says AI will soon surpass humans in coding

29-Aug-2025

.

Large language models (LLMs) such as GPT-4, Claude 3.5 Sonnet, and Gemini 2.5 Pro are already writing and debugging code like professional coders. And what began as AI assistants for programmers is now raising concerns of replacing human coders altogether. A senior AI researcher at Alibaba believes that this future is not very far away.

Binyuan Hui, a staff research scientist at Alibaba’s Qwen team, recently shared his thoughts on X (formerly Twitter), arguing that “LLMs will inevitably surpass humans in coding.” Not only is AI mirroring how humans learn, but it is also getting so advanced that it is even outcoding humans.

In his post, Hui explains that human coders typically go through two stages. The first, according to him, is memorisation and imitation. Where the coders learn syntax, study examples, and replicate good projects. The second stage is trial and error. In this stage the programmers write code, run it, fix bugs, and improve through feedback.

Now with LLMs getting smarter, Hui argues, that the AI is also following the human path of learning. Pretraining is allowing the models to absorb and compress vast amounts of code that is in fact far beyond what any human could possibly memorise. “Reinforcement learning (RL) aligns the model with execution through feedback, but at a speed and scale humans cannot match, with millions of rollouts completed in a short time,” writes Hui.