The limitations of ChatGPT or why isn't it good enough for the MEP industry

Following OpenAI's announcement of new models and developer products, Pelles AI introduced specialized ChatGPT tools for MEP subcontractors. However, general-purpose AI tools fall short for construction professionals — domain-specific alternatives are needed.

The Top 5 Limitations of GPT for Construction

1. GPT struggles with specialized topics and industry-specific context

Despite extensive information access, ChatGPT cannot effectively answer context-specific or niche technical questions.

2. GPT Hallucinates Facts

The system tends to fabricate information to complete answers. OpenAI has acknowledged this "tendency to provide incorrect answers."

3. GPT lacks common sense

While responses sound human-like, the system generates nonsensical outputs when practical reasoning is required.

4. GPT doesn't have internet access or real-time data

Knowledge cutoff limits the system, with no access to current information or pricing.

5. GPT's answers are not explainable and users can't work with the document

Responses lack underlying assumptions or source documentation, making verification impossible.

Real-Life Examples of GPT's Limitations

Question 1: Ceiling Diffuser Count

When presented with a mechanical works bid package for a 4,000-square-foot NY restaurant renovation, GPT provided fabricated quantities despite identifying diffuser types correctly. The quantities were wrong because GPT cannot count drawings or understand that the number of diffusers isn't shown on the schedule and has to be counted on the drawing.

Question 2: Equipment Pricing

When asked for current pricing on a Versatec 700-048 water-cooled DX system, GPT generated lengthy responses without providing actual data, demonstrating its lack of real-time market information.

Question 3: Equipment Type Identification

GPT failed to distinguish between major systems and components, lacking the contextual understanding needed to identify MAU and PCU units while mistaking unit designations as model numbers.

Conclusion

While GPT delivers promising results for MEP professionals, the real test lies in tailoring AI models to provide customized, high-value solutions specific to the MEP sector. This represents an exciting yet unmastered frontier.