Generative AI Accuracy: Science or Art?

Seed any talented, experienced Founder with $1Billion at day zero of their mission, and they’re likely to deliver something seismically impressive 7 years later.

And so it goes with Sam Altman. When ChatGPT dropped, it went viral. Spellbinding 100 million people in less than 2 months.

Looking past the headlines, what I find fascinating is the direction OpenAI took in allowing ChatGPT’s Users free-form prompt entry.

Free-form entry is when your computer interface allows you to write (or draw) inputs freely.

In a Generative AI input context, it’s equivalent to handing over the keys of every car in your dealership to prospective buyers for a test drive without a shred of identification or collateral from them, and hoping they return.

It’s a big risk. Because you simply don’t know what they’re going to do.

Sam & Co. took that risk. Letting their users write anything that occurred to them, and trusting that the system they’d built would be able to generate some kind of response. Kudos to them.

Truth Be Damned.

But there’s a little secret that swooning media headlines don’t speak of.

Aside from enormous capital resources, Sam had another luxury afforded to him. That the degree of accuracy of ChatGPT’s responses, wasn’t a priority.

Safe in the knowledge that improvement in accuracy would come with later iterations of their GPT framework, what mattered more was that ChatGPT generate a response. Any response would do, so long as it were contextually and syntactically correct. Even if it weren’t entirely true.

In other words the priority for ChatGPT was to be output-oriented, and that accuracy would be improved down the line.

Software As A Service. Accuracy As An Obsession.

There are less funded Founders that do not have such resources and the learning curve it affords.

I’m one of them.

Having set out in 2019 to develop a themed language model, we were cognisant of the need to deliver accurate output, or else we’d fail to launch.

Our entire focus became outcome-oriented. We needed to ensure that our Generative AI would deliver content of value specific to our target Users’ use case. is a real estate themed language model.

Written in a combination of languages, overwrite is trained through reinforcement learning and rule-based action, using a proprietary corpus of over 2Billion scale parameters to generate syntactically, contextually and factually correct real estate marketing content.

Used by the real estate industry to instantly generate unique and engaging property listings. Content that can neither be generic nor factually misleading, for search optimisation and legal reasons respectively.

People are increasingly understanding the value of Generative AI accuracy.

Input First. Output After. Sounds Easy Enough.

To deliver that absolute accuracy, we needed to guide our Users through the input process. We couldn’t expect them to know what to prompt the system with. How could they? Before ChatGPT’s release the very concepts of Generative AI and Prompt Engineering were known only to those within the industry. Certainly not to our real estate marketing audience.

So we developed a prompt input system of our own. One that takes Users through a dynamically responsive journey. Meandering. Adapting prompts and predicting inputs as it goes. Never risking human error, insufficient inputs or hallucinatory outcomes.

We knew at the time that consumer-friendly prompt engineering would one day become critical to the widespread adoption of AI tools. (Fun Fact: Today Prompt Engineer. salaries are reaching $400K)

That meant finding the Goldilocks Zone between the AI science, and the very human art of communication.

Designed for specific use cases, Themed Language Models like overwrite do not need to be all things for all people. They can be fit-for-purpose more cost-effectively, and user-efficiently than their large language model cousins. Now that frameworks make generative AI model development easy and fast for enterprises; developers must remember to strike that UX balance between what’s good enough for them, and what’s expected from their Users.

Ayman is recognised as one of MENA’s leading Generative AI pioneers.

He is the Founder and CEO of, the MENA region’s pioneering Real Estate Themed Generative AI, an NVIDIA Inception Program partner.

With +20years of corporate and entrepreneurial experience, he regularly features on thought leadership panels, blogs, vlogs and radio segments, sharing his often inspirational opinions about Artificial Intelligence and entrepreneurship. is a pioneering Themed Generative AI, creating engagement-oriented content for the real estate industry. 

We create the marketing content that powers the real estate industries of the UAE, KSA, Egypt and Lebanon.

For informative news and views on the world of real estate, proptech and AI, follow overwrite on Instagram and LinkedIn, and keep up-to-date with our weekly NewsBites blog.

overwrite | real estate content creation, reimagined.

Notify of
Inline Feedbacks
View all comments