Where Does Knowledge Come From in the Age of AI?

Throughout my career, I've often needed instant access to information, from incorrectly configuring a Restart=on-failure setting in a systemd unit to an unnecessarily large PostgreSQL connection pool. For years, like most developers and system administrators, my first stop was always the same: Stack Overflow. However, the question of Where Does Knowledge Come From in the Age of AI has recently become one of the most perplexing questions on my mind.

This shift isn't just about moving from one tool to another; it signifies a fundamental change in how we view the nature of information. When I encounter a performance problem in a production ERP system, what used to be hours of digging through Stack Overflow now translates into an AI chat. But does this new method always provide accurate and reliable results? That's where the real issue begins.

The Evolution of Information Hunting: Stack Overflow and Before

I remember the old days, when the internet wasn't so widespread, knowledge came from books, manuals, or even experienced colleagues in IRC channels. I recall spending days reading documents for a VLAN segmentation issue, or sifting through dozens of pages of RFCs for a routing table. Stack Overflow fundamentally changed this process.

Suddenly, the answer to almost every error, every configuration problem I encountered, was just a few clicks away. A Nginx reverse proxy setting, a Redis cache strategy, or service orchestration with Docker Compose... Thanks to quick and precise answers, I solved problems much faster. This was an incredible boon for someone like me, who works in the field and is focused on practical results.

ℹ️ The Value of Stack Overflow

Stack Overflow continues to be an invaluable resource for specific code snippets, error messages, and common configuration issues. Especially when you know exactly what you're looking for, its speed is hard to match.

The Impact of Artificial Intelligence on the Search for Knowledge

Since AI models entered our lives, our way of accessing information has been revolutionized once again. Now, instead of just searching with keywords, I can describe my problem in natural language, as if talking to a human. Asking about different vacuum strategies for a PostgreSQL WAL bloat problem or learning what to consider when setting a cgroup memory.high limit has become much more fluid.

Initially, I struggled a lot with prompt engineering to get the right answer, and sometimes I even fell victim to AI's "hallucinations." Once, a code snippet I got from AI for JWT token validation for the backend of my own side product contained a logical error that could lead to a security vulnerability. Fortunately, thanks to my audit processes, I caught the error before it went into production.

# Faulty JWT validation example suggested by AI (simplified)
# Remember this code is faulty and should not be used in production!
def validate_token_ai_suggestion(token):
    try:
        # AI might sometimes omit or incompletely write this part
        # For example, algorithm check, issuer check might be missing.
        decoded = jwt.decode(token, "SECRET_KEY", algorithms=["HS256"])
        return decoded
    except jwt.ExpiredSignatureError:
        print("Token has expired!")
        return None
    except jwt.InvalidTokenError:
        print("Invalid token!")
        return None

Such situations taught me to question the information provided by AI and always cross-reference it with human confirmation. AI is a great assistant for synthesizing information and gathering it from various sources, but I still make the final decision and bear the responsibility.

The Value and Context of Information: Why Human Experience Still Matters?

The answer to the question Where Does Knowledge Come From in the Age of AI no longer comes from a single source. While AI offers us a fast and broad spectrum of information, the "context" and "experience" of that information are still hidden in the human factor. You can get general recommendations from AI to solve a VLAN tagging mess, but AI cannot know the specific switch model, firmware version, and more importantly, the long-standing network architecture decisions of that organization.

When developing an ERP for a manufacturing company, working on an event-sourcing architecture, AI could tell me all the theoretical benefits and challenges of this architecture. However, in the real world, the practical difficulties of ensuring idempotency or implementing the transaction outbox pattern, and how they aligned with the company's existing operational flow, could not be solved with AI's "general" knowledge. What was critical there was my 20 years of field experience, my ability to understand that company's culture and operations.

When working with RAG (Retrieval-Augmented Generation) patterns in AI application architecture, AI having its own internal database and feeding from that data provides a great advantage. However, no matter how much prompt engineering I do, when choosing a PostgreSQL partition strategy, AI alone cannot optimally determine how frequently data is written and read, how often historical data is accessed, and how this decision will reflect on vacuum monitoring. This requires someone like me to be able to say, "We would have done X, but we chose Z because of Y."

New Knowledge Sources and the Future

So, Where Does Knowledge Come From in the Age of AI? For me, it's now a multi-layered process. In the first step, I use AI as a tool for quick preliminary research, synthesizing different approaches, and even creating new regex patterns for fail2ban. However, for critical decisions, I always delve deeper:

• Official Documentation: Always the most reliable source. Especially for security topics like Linux kernel module blacklist, it's essential to follow CVEs and read the relevant documents.
• Reliable Blogs and Communities: I follow blogs that offer in-depth analyses based on real experiences and active communities.
• Human Network: Conversations with colleagues and industry experts I've gained over the years often provide "gray area" information that I can't get from AI.
• My Own Experience: My most valuable source of information. Last month, when I typed sleep 360 and got OOM-killed by a systemd cgroup limit, I examined my own journald logs before asking AI about the error and solved the problem with a polling-wait mechanism. These "learn by doing wrong" experiences are lessons no AI can teach.

💡 Validating AI Output

Make it a habit to always validate technical information you receive from AI. Especially in security, performance, and critical infrastructure settings, cross-checking with official documentation and your own experience is vital.

Conclusion: Wisdom Still Resides in Humans

Artificial intelligence is a powerful tool that accelerates and simplifies our access to information. However, we still determine the difference between information and wisdom. AI can offer me many suggestions on how to optimize a PostgreSQL query, but making the most accurate decision by considering the criticality of that query in the workflow, possible deadlock scenarios, and the company's overall SLAs is the responsibility of a 20-year technology professional like me.

So, the answer to the question Where Does Knowledge Come From in the Age of AI is: it comes from everywhere. But making sense of this information, putting it into context, and transforming it into correct decisions is still our job. AI can be an assistant, a guide; but the captain's seat is still ours.

So, how do you access knowledge in this new era? What was the most valuable information you got from AI, or what was its most misleading answer? I'd love for you to share it with me in the comments.