How Ordinary Programmers Can Transform Into AI Engineers: Real Success Stories

With graduate salaries reaching 500 k and overseas AI talent starting at 1 M, the article asks whether ordinary programmers should become AI engineers and shows three typical Zhihu cases of successful transitions.

Case 1

Step 1: Review Linear Algebra – the author used MIT’s open linear‑algebra course and shared a GitHub notes link.

Step 2: Learn Machine‑Learning Algorithms – followed Stanford’s CS229 (Andrew Ng) videos, covering linear models, SVM, clustering, EM, PCA/ICA, learning theory, and Markov models.

Step 3: Implement Algorithms in Code – used Coursera’s ML course to code models in MATLAB, then moved to Python.

Step 4: Build a Complete Model – studied CS231n (CNN) videos, completed Jupyter‑based assignments using NumPy, SciPy, and Matplotlib.

Future plans include Geoffrey Hinton’s neural‑network course and Stanford’s CS224d for NLP.

Case 2

The author, originally strong in ACM algorithms, shifted to data mining and now pursues a data‑science career. Essential skills include C++, Java, Python (web crawling, numerical computing, visualization) and solid foundations in linear algebra, calculus, and probability.

Key statistical tools: correlation analysis, regression, clustering, distribution modeling, evaluation metrics, and A/B testing. Recommended reading: Li Hang’s *Statistical Learning Methods*.

Feature‑engineering checklist: usability assessment, cleaning, sampling, scaling, encoding, transformations, dimensionality reduction (PCA, LDA, SVD), feature selection (filter, wrapper, embedded), derived features, and monitoring.

Suggested platforms: Spark, Caffe, TensorFlow.

Case 3

A programmer from a non‑IT field leveraged domain expertise and large‑scale data, treating deep learning as a tool. Steps include installing TensorFlow, studying the MNIST example, learning basic neural‑network theory, and exploring frameworks like Theano.

Recommended resources: Andrew Ng’s ML Coursera, Karpathy’s char‑RNN blog, neural‑style projects, and Keras documentation.

Advanced reading includes original CNN, RNN/LSTM, reinforcement‑learning papers, and DeepMind Nature articles. The author also shares links to Kaggle, GitHub repositories, and recent pre‑print papers.

Final advice emphasizes continuous coding, staying updated with papers and tools, and applying AI techniques to one’s own projects.