Two Major Career Paths in AI Large Language Models: Engineering Implementation vs. Algorithm Research

The LLM Job Market Is Diverging

As AI large language model (LLM) technology advances rapidly, more and more programmers are looking to transition into this field. But what does the LLM job market actually look like? Which roles are right for you? What kind of education and skills do you need?

Recently, Bilibili content creator "码士集团" (Mashi Group) conducted an in-depth analysis of the AI LLM job market during a livestream. The core takeaway was crystal clear: The LLM job market will split into just two major directions — engineering implementation and algorithm research. All related positions will fall into one of these two branches, and they differ significantly in education requirements, skill thresholds, and career prospects.

Path One: LLM Engineering Implementation — The Main Battlefield for Most People

What Is Engineering Implementation?

LLM engineering implementation, in simple terms, means leveraging existing foundation models (such as DeepSeek, OpenAI, etc.) and combining them with enterprise business needs for secondary development and deployment. Specific work includes:

• AI Agent Development: Building intelligent agent applications with autonomous decision-making capabilities based on LLMs
• RAG (Retrieval-Augmented Generation) Development: Integrating enterprise knowledge bases with LLMs to improve answer accuracy and domain expertise
• Fine-tuning: Adjusting foundation model parameters for specific business scenarios
• Multimodal Applications: Calling models like YOLO for image processing and combining them with LLMs for multimodal interaction
• Knowledge Graph Integration: Combining knowledge graphs with RAG to create advanced applications like Graph RAG

Roles and Requirements

This path covers positions such as: LLM Application Engineer, LLM Application Algorithm Engineer, AI Agent Engineer, LLM Engineer, and more. The minimum education requirement is a bachelor's degree, making it relatively accessible and suitable for most developers with a programming background.

However, there's a critical caveat: Don't assume that the engineering implementation path means you can skip learning algorithms. As the content creator emphasized, this is a common misconception. Even when working at the application layer, a foundational understanding of machine learning and deep learning algorithms is still essential. Claims that "you don't need to understand algorithms to build applications" are a serious misrepresentation of this industry.

Future Trends

In terms of job volume, engineering implementation will undoubtedly be the direction with the highest demand. As more and more enterprises integrate AI LLMs into their business processes, the need for application-layer development talent will continue to grow.

But the flip side is: More and more people will flood into this direction, making intense competition inevitable. As for when that competition will peak — whether in two to three years or later — it's impossible to predict precisely. But one thing is certain: this path will eventually transition from a blue ocean to a red ocean.

Path Two: LLM Algorithm Research — High Barrier, High Reward

What Is Algorithm Research?

The LLM algorithm research path is primarily responsible for innovation and optimization at the foundational algorithm level, including:

• Writing and optimizing operators
• Optimizing LLM training processes
• Improving LLM inference efficiency
• Exploring new network architectures and training paradigms

Currently, this path mainly focuses on algorithm research in the AIGC domain, but it may expand into cutting-edge areas like embodied intelligence and world models in the future.

Roles and Requirements

Positions in this path include: LLM Algorithm Engineer, LLM Algorithm Expert, Multimodal Algorithm Engineer, and more.

The requirements are very clear and strict:

• Minimum education: Master's degree from a top-tier (985) university
• Major requirement: Must be in computer science or mathematics (formally trained)
• Publication requirement: Must have papers published at top conferences or journals — ordinary journals or ghostwritten papers don't count

A Counterintuitive Fact

Despite the extremely high barrier to entry, the content creator pointed out an interesting phenomenon: The algorithm research path is actually the easiest direction to find a job in right now — bar none. The reason is simple — very few people can meet these hard requirements. The market is desperately short of talent, with severely insufficient supply, which paradoxically makes the job market in this direction the most favorable.

Foundational Algorithm Knowledge: The "Constant" in AI

Here's a noteworthy detail: although hot topics and directions in AI have constantly shifted over the past decade-plus — from image recognition and object detection to AIGC — the underlying machine learning and deep learning algorithm frameworks have not fundamentally changed.

From the emergence of the YOLO model around 2017 to today's large language models, the core remains the same neural network ecosystem: convolutional neural networks, deep neural networks, the Transformer architecture, and so on. Changes are mostly reflected in network topology and training strategies, not in an overhaul of the underlying mathematical principles.

This means: if you build a solid foundation in core algorithms, your competitive edge will endure even if the application direction of AI shifts in the future.

The Programmer's Lifelong Learning Destiny

After analyzing both major paths, the content creator shared a deeper career insight: The programming profession is destined to be one where you learn for as long as you work.

He used his own experience as an example — starting with front-end development and continuously pivoting as the industry evolved. No one can ride a single skill for an entire career. If you stop learning, you'll be phased out by the market in roughly three years. This isn't unique to the LLM field — it's a fundamental law of the entire IT industry.

Even if you enter a "high-end" direction like algorithm research, coasting after landing a job means facing obsolescence in three years just the same. The pace of technological iteration doesn't slow down based on your education level.

Practical Advice for Those Looking to Enter the Field

Based on the analysis above, here are some recommendations for programmers looking to enter the AI LLM space:

1. Know your positioning: Choose the right path based on your educational background and professional foundation. If you have a bachelor's degree, prioritize the engineering implementation path. If you have a master's from a top-tier university with published papers, you can aim for the algorithm research path.

2. Don't neglect algorithm fundamentals: Even on the application path, foundational knowledge of machine learning and deep learning is a required course. Otherwise, you'll quickly hit a ceiling in technical depth.

3. Maintain continuous learning: The only certainty in this industry is change. Regardless of which path you're on, stay sensitive to new technologies and keep your learning ability sharp.

4. Prepare mentally: Intense competition in the engineering implementation path is coming sooner or later. Building differentiated competitive advantages early — such as domain expertise in specific industries or full-stack capabilities — is the long-term strategy.

5. Follow trends rather than fight them: Just as front-end developers now need full-stack capabilities, the skill boundaries in AI are constantly expanding. Flexible adaptation is the key to survival.