Can you trust artificial intelligence?

Walter Damián Cabrera, founder of LIA Engine · October 29, 2026

How reliable are AI answers?

It depends a lot on what you ask. For writing, summarizing or brainstorming, it usually works very well. Problems appear when the answer has to be exact:

• Calculations and totals: a miscalculated VAT or a sum that doesn't add up.
• Dates: shifted due dates or dates that don't exist, like February 31.
• Data taken from documents: a number misread from a photo or a PDF.
• Sources: quotes or links that don't exist.
• Long spreadsheets: a row left out of a formula.

The problem isn't just the error. It's that it doesn't warn you: a wrong total looks exactly like a correct one.

Two digits swapped. Nobody notices at a glance, and the invoice goes out wrong.

Is AI 100% reliable?

No. No AI is 100% reliable, not even the newest ones. They keep improving, but they still get things wrong from time to time, and you never know in advance in which answer. That's why the useful question isn't whether AI is perfect, but how to check what it answers before the error costs money.

Why does AI make up data? Hallucinations

A language model writes the answer that seems most likely, word by word, based on what it learned from its training data and what it has already written. It doesn't check the result. If it doesn't have the data, if the source document is hard to read or if the calculation is long, it may fill the gap with something that sounds right. That's a hallucination: a confident statement that isn't true.

Hallucinations became a well-known topic in 2022, with the arrival of ChatGPT and other LLMs. One of the most cited examples is precisely an AI that is asked for a financial statement and answers with made-up figures. It doesn't do it on purpose: it simply has no way of its own to know whether its number is right. We explain this in detail in What are AI hallucinations? · More about hallucinations on Wikipedia.

Which AI is the most reliable?

It depends on the task, and the ranking changes from month to month. One AI can be excellent at writing and fail at calculations, and another the other way around. General comparisons are useful as a guide, but the most honest way to know is to measure it with your own tasks.

At LIA Engine we built a ranking of the AIs that certify most from customers' real validations, not from tests paid for by anyone.

Which is more reliable, Google's AI or ChatGPT?

Both are right a lot and both can be wrong. Which one works better depends on the task and on the current version. For work with numbers, which one you choose matters less than having someone check the answer before it's used.

Is it dangerous to use ChatGPT or another AI for work?

The most common risk isn't dramatic: it's a wrong number that nobody checks and that ends up on an invoice, a tax computation, a quote or a report for a customer. The other thing to watch is privacy: before pasting customer data, read what the provider does with what you send.

So, why not use AI?

On the contrary: use it, because it saves a huge amount of time. But where it matters, check. Five habits that help:

1. Redo the key calculations: VAT, totals, interest.
2. Make sure the dates exist and make sense.
3. Compare the total with the sum of the parts.
4. If it cites a source, open it.
5. Don't paste sensitive data without checking the provider's privacy policy.

Doing it by hand works, but it takes time, and when there are hundreds of answers a day it's no longer enough.

How LIA Engine solves it

LIA Engine sits between your systems and the AI you already use. Your AI answers, LIA checks the answer against your data and the rules of the request, corrects what can be calculated and delivers a signed certificate that anyone can verify. What can't be checked, it doesn't certify, and it tells you why.

In our internal test bench (October 2026, with sample data), the same AIs scored 25 to 100 points more correct answers depending on the task, with no wrong number certified. See the results.