This site uses cookies to provide you with a great user experience. By using the site, you accept our use of cookies.

13 Mar 2024 • Tom Haley

Navigating the new world: a QS's guide to generative AI tools

In the last couple of articles, I provided some information which summarised my understanding as to how generative AI (large language models) work and the risks and limitations of these tools.

This week’s article continues our generative AI mini-series, and I focus on my experience of using the generative AI tools. The last couple articles have been heavier reads than usual, so I have deliberately kept this quite light-hearted and look at it from a practical QS perspective.

It’s fair to say that there are hundreds of generative AI tools available so, in the interests of your time and my sanity, I have focused on the three that I think QS’s may use the most both now and in the future. These are Chat GPT, Microsoft Copilot and Adobe AI Assistant.

Chat GPT

The original. Whilst it may not be the first generative AI, it will always be remembered that way, like Facebook is to social media.

I quite like the Chat-GPT interface. It is easy to use, and it isn’t over complicated. You type in your question, upload your files, or explore what is on your mind at that point. The information you get is generally along the right lines and, with further prompts, you tend to either get what you need or enough of what you need to keep yourself moving with whatever task you are on with. There have only been one or two occasions where ChatGPT just hasn’t got it and I have had to move on without satisfaction.

I have learned, through reading about how generative AI works, and my own experience, that the question you ask is vitally important. I suppose the same can be said of working in business and life. You think you have been uber clear about the question yet, despite asking again and again, you just can’t get satisfaction in the answer you are given.

The learning process I have been through it to really challenge myself on the question I am asking. Is there enough detail? Is there too much detail? Have I misled the AI by talking about something a lot which wasn’t that important to what I wanted and talking too little about the thing that was actually really important.

One final point regarding ChatGPT, please never, ever copy and paste answers directly from ChatGPT into an email or a letter. Nobody in life, or in the business world, that I can recall coming across, communicates like an AI. I can spot it a mile off, and so can others. You may think the extra pazazz in your language makes you look smart, but it doesn’t. it achieves the opposite.

Use generative AI for inspiration to give you some structure or for inspiration when you have a mental block, but make sure everything you do goes out in your own voice. It matters, I promise. And I also promise that most people want to hear from you as your authentic self, not from a bot.

Microsoft Copilot

The eagerly anticipated Microsoft generative AI. I watched the videos which were like movie trailers (well, not quite, but let’s continue with the build-up). The opportunities and possibilities going through my mind were insane. This has the potential to exponentially improve my work performance and productivity, and those around me too. This could be incredible.

Then came the launch. I updated Microsoft and waited the customary 4-5 minutes for the machine to whirr. It popped up and, like a kid at Christmas switching on their new games console for the first time, I was bright eyed thinking what will my first question be.

I lived up to my QS reputation and went for something wild, that’s right, I asked Copilot to “generate tips for effective time management”. It was a suggested question, it was a question that had been on my mind recently, so why not. And the answer arrived. 18 detailed points with some suggested further reading. The irony of my question and the detail of the answer were not lost on me.

I quickly realised that Copilot has a long way to go. It is a bit like Bing standing up rather than lying down. Sometimes you get some wheat maybe 1 in 10 hits, but it doesn’t seem worth sorting the 9 out of 10 chaff you have to suffer to get to it, so you ditch Bing and go back to using Google Chrome.

I definitely think there is potential here. I hoped for a tool which would sit alongside me whilst using Microsoft applications (remember that little paperclip? Something like that) which detected my user level and what I was trying to achieve and could give me targeted ideas about how I could improve what I was doing.

That is where the opportunity is for me. Microsoft’s development of its applications is happening every day, but most people, me included, tend to stay within their comfort zone. Encouraging people out of that and showing them the possibilities could unleash so much untapped improvement potential.

I haven’t given up on Copilot and expect it will seriously up its game in the coming months and years. For now though the icon sits in the bottom right of my screen gathering dust.

Abode AI Assistant

The one I knew I needed but wasn’t expecting. When that colourful little button first popped up in the top right of the Adobe application, my eyes lit up. It was so alluring – starry, multi-coloured, perfectly sized and just ready to be hit.

Again, keeping with my wild QS theme, I had a draft adjudication Referral in my inbox that was on my action list to check, so I converted it to pdf and thought I would ask the AI assistant some questions. The excitement was palpable.

I asked it to give me an exec summary, which it did. I knew the dispute and the issues, and I would say the exec summary was about 60% good. I wouldn’t copy and paste what it gave me and use it blind (see previous tip) but it would definitely give you a solid head start.

I asked the AI, cheekily, whether the adjudicator would decide in our favour. The AI, very wisely, suggested I seek legal advice. The answer was polite, firm and, to be fair, very correct.

Fun aside, I see massive opportunities here. How often do you have to deal with voluminous pdf’s or multiple pdfs with hundreds, if not thousands, of pages of information. Think contracts or tender documents. It’s 5pm and you have been asked to review something before it is needed by 9am in the morning. Hopefully we will see the back of those poor work practices, but it happens.

This tool will allow you to navigate pdf documents, get what you need and summarise issues that can be scattered across documents. An example? Design responsibility. Could something like this tool enable you to pull together a mass of pdfs included in the contract and summarise this? Putting those pesky clauses buried away in specs alongside the rest of the terms to give you a more complete understanding in a very quick time. Even if it’s 60% right, it gives you one hell of a good start.

The possibilities in my mind are endless, however the AI assistant does currently have limitations. The tool only works with documents which are 25MB in size or 120 pages long. I expect those limitation will only be temporary. Keep your eye on this tool, it will be game changing for us QS’s.

Final reflections

There is so much changing, every day, when it comes to generative AI tools so hopefully this gives you some insights into how I have found the process of user the tools. If you are a non-user, I hope this encourages you to give a go and if you are a regular user, I hope this inspires you to push the boundaries and possibilities further.

In next week’s article, I will take a bigger picture view about the possibilities generative AI might present to the quantity surveying profession. This thinking will not be constrained by how the tools operate currently, but will seek to push boundaries with a focus on what I think we need to be more productive or produce better quality work.

Keep your eyes peeled.

Back to insights



We publish insights through our LinkedIn newsletter, titled “The science of Quantik”, which are light bites of information covering news and insights relating to the construction industry and quantity surveying.