Lesson Content

Six core limitation categories.

1. Current information AI knowledge ends at the training cutoff. Events, product releases, regulatory changes, scientific publications, and market conditions after that date are unknown to the model. For any time-sensitive query, verify currency through live sources.

2. Reliable factual accuracy AI generates plausible text, not verified facts. Specific figures, citations, dates, names, and technical details can be wrong – sometimes confidently wrong. The generation process does not include a fact-check step. For any specific factual claim you will act on, verify.

3. Genuine reasoning and judgment AI applies learned patterns, not genuine reasoning. For familiar problem structures, pattern application looks like reasoning. For truly novel problems – ones with no good analogues in training data – AI will pattern-match to something adjacent, sometimes incorrectly. High-stakes, novel judgment calls belong to human expertise.

4. Self-knowledge and calibration AI systems are poorly calibrated about their own uncertainty. They do not reliably know when they know something well versus when they are confabulating. A wrong answer often comes with the same confidence as a correct one. Do not use AI confidence as a signal of accuracy.

5. Real-world physical and sensory experience AI has never experienced the physical world. It has read about physical tasks, materials, sensations, and environments extensively – but it has no direct experience. Tasks requiring practical physical judgment – construction decisions, material assessment, interpersonal physical dynamics – are outside its reliable range.

6. Persistent memory across conversations Each conversation starts fresh. AI does not remember prior conversations, prior preferences, prior errors corrected, or prior instructions given. Without Projects and Custom Instructions to load context, each new conversation is a blank slate.

Why limitations matter for practice.

Knowing where AI falls short does not reduce its usefulness – it makes you a better user by helping you recognize when to rely on AI output and when to verify, seek human expertise, or take a different approach. The professionals who get the most from AI tools are not the ones who trust them most; they are the ones who understand them most accurately.