Why Energy Permitting Requires More Than ChatGPT

When we tell people that we’re using AI to streamline and accelerate permitting for utility scale energy projects, often one of the first questions that we’re asked is, “Why can’t an energy developer just use Chat GPT?  Why do they need Permeta?”

That’s a great question.  Publicly available LLMs like ChatGPT and Claude are great at many things- editing emails, travel itinerary planning, offering headline options for a blog post (thanks Chat GPT for this headline).  However, they’re not built to handle 1,000+ page technical documents filled with precise, confidential data that must consistently meet strict and constantly evolving federal, state and local regulatory requirements.

Here’s why.  

1- Size.

Utility scale energy permit applications are massive, 500 to 5,000+ pages.  These documents require coordinating engineering studies, environmental data, maps, models and regulatory filings across multiple experts and jurisdictions.  Publicly available LLMs are not designed to reliably work across that volume of structured, interdependent information.

2- Security

Permitting data is highly sensitive– including site details, infrastructure plans, and proprietary analysis.  Uploading that data into a public LLM without strict controls can expose the data in ways that many developers are not comfortable with. For most energy companies, that alone is a non-starter.

3- Hallucinations

LLMs are trained to generate human-like text, not accurate text.  The models are trained to predict what sounds right, not necessarily what is right. That’s why they can occasionally produce confident but incorrect outputs, including fabricated data or sources.  

Essentially, LLMs were trained just as many of us were trained to take tests in school.  If you’re taking a multiple choice test and you don’t know the answer to a question, most of us were told to guess and pick an answer vs. leaving a question without a response.  If you guess, you have a decent chance that you’ll randomly guess the correct answer.  If you leave the multiple choice question blank, you’ll certainly get a zero on that question.  When LLMs don’t know the answer, rather than admit they don’t know and leave the field blank, they’ll make up a response that looks and sounds plausible.  

So, what does Permeta do differently?  Many things.  We build purpose-built AI systems for permitting; our system is designed to handle large-scale documents, enforce structure, and ensure outputs align with regulatory requirements.  This includes approaches like neuro-symbolic AI, which help eliminate hallucinations and improve reliability. More on neurosymbolic AI next week.  Stay tuned. 

In the meantime, keep using Chat GPT to draft emails.  But leave streamlining energy permit applications to Permeta.  

If you’d like to read more about LLMs and how they work, this piece by Andreas Stoffelbauer, former data scientist at Microsoft, is especially helpful.

https://medium.com/data-science-at-microsoft/how-large-language-models-work-91c362f5b78f