Does Context Still Matter In the Large AI Era?

It’s given that generative AI is getting bigger and better. GPT4 is believed to boost some 1.7 trillion parameters, and its successor is likely to exceed that. It will know “everything”, right?

Perhaps the answer is yes for general knowledge like the stuff you can find on Wikipedia. However, if you prompt the AI for a specific piece of information that it has not seen when trained, it will hallucinate (with confidence!)

In other words, if you need to know a specific word, say in Arabic, would you rather ask a renowned professor in English literature in Cambridge, or a young lad from the Middle East?

If we think about this problem from a real estate lens, to what extent can you rely on ChatGPT to give you tailored advice on where to buy or invest in properties in the UK?

The model size doesn’t matter. What really matters the most is the context. Adding context to the prompt is a process known as RAG, or Retrieval Augmented Generation.

At Rexsmart, we have built a large database to train our bespoke property valuation AI (AccuVal). The database contains data on properties, prices, locations, amenities, as well as stats that we generate from the data, such as the median and average square meter price for every postcode and neighbourhood.

I am thrilled to announce that we have started to build a specialised AI assistant to answer property related questions. Our goal is to provide a neutral “advice” to help buyers and sellers with their housing journey. The current alternative is to speak to your local estate agent!

10-05-2024