Let’s be Superhuman.

I’m going for the numbered list format this week:

1. I said last summer that I wouldn’t mention AI again. Well, all that’s changed these past few weeks. I’ve been deep in ChatGPT and discovered other versions that generate high-quality images directly from my hand-drawn sketches.

2. I’ve been working ChatGPT hard. Yesterday, I blew its brains out analysing data from the Office of National Statistics on construction costs for the past ten years. It’s not faultless; you must find the right way to prompt it, but a key component of the process is to prime or train it.

3. I was up last night attending a live training session. Your key to getting the most out of any Large Language Model (LLM), of which ChatGPT is just one, is to train it and treat it like a new employee. Use each individual Chat as an expert colleague whom you train specifically for the task; for example, I now have one who is an expert at analysing construction data.

4. I must ask my new colleague to summarise what he knows regularly. That way, it doesn’t forget—something to do with tokens. Your goal in training your new assistant is to have it come back and ask you questions. That way, you’ve got it thinking about the specialist area it’s being trained for.

5. Generally, never bother with Google for any research again. Want a recipe? Ask your LLM. Indeed, train an expert assistant to help you only with recipes. You only have spaghetti, an aubergine, garlic and a tomato? Tell it that, and it will give you a recipe or five.

6. Before you start, though, prime it. Tell it what you want it to help you with, and give it some context to the problems and tasks you want it to help you with. This is the training bit; you’ll get better results if you work it this way.

7. I’ve been teaching first-year engineering students for the past two weeks. In the case of the Chemical Engineers, they are designing toilet systems in rural Africa. They have just started and are in the research and concept development stage. Out of nearly 600, not one I’ve spoken to is using ChatGPT. Madness. You can ask it to help with the base research. What are the existing types of waterless toilet systems? You get a list. Which of these is suitable for rural Africa? You get a refined list. Pick one on the list, a UDDT – urine-diverting dry toilet in case you’re wondering – and ask for more details. In minutes, you’re into your subject.

8. Asking it to summarise long articles, either from a PDF or webpage, is a brilliant use. I’ve asked it to read, summarise and compare multiple articles on the Building Safety Act and I’ve shown students how to do the same for their projects. Now, it depends on the context of your research. If you’re doing a PhD, you’d clearly deal with reading and summarising large amounts of documents differently to a two-week design project. So, as with all tools, it’s how you use it appropriately for the important task. You’d not use a putter out of a bunker.

9. Brainstorming ideas is a brilliant use of your AI helper. You can go from blank paper to ten suggestions in seconds. Drill down, polish the outputs, and don’t just take the first answer. Read it, digest it, and ask more questions. There is a view that somehow it’s cheating. It is if you copy and paste what it gives you and pass it off as your work. It’s not if you want to improve your work and when you are transparent about your process.

10. Then, ask it to create an image representing one or a few ideas. It is astonishingly good at this. Again, don’t just take the first version. Ask it a few times with detailed prompts. What pops out is not something any engineer or designer would actually use, but it creates a series of sketches that get you thinking.

12. A colleague at the university asked me about his loft design and said, “But aren’t you worried AI is going to be able to design loft conversions?” Well, it will be able to do it. Not much point hiding from it. “We need to be training this lot how to use it,” I say, pointing at the roomful of future engineers. “Teach them to be creative with it, how it will make them better engineers and designers. They need to know this right now, so their brains don’t turn into a fluffy Tick-Tok mush.” I actually think this might be the chance we have to be super-humans. I’m excited. Better get on with it.

13. Can’t be bothered to read this? Paste it into GPT and ask it to summarise it in 100 words.

This week’s web links include food trends for 2024, a ski slope and ideas for multi-generational living in terrace houses.

Feel free to let me know if you have any comments or suggestions. You will always find me at carl@carlarchitect.co.uk.

This Week’s Links:

What exactly is a UDDT?

Get yourself an OpenAI account and try out ChapGPT. If you want to use it seriously, the monthly $20 fee is well worth it.

And then you could watch this good tutorial on ChatGPT for beginners. Be superhuman by next Wednesday, haha.

And then do the training course I did last night with Jordan from Everyday AI – also a good podcast.

Food trends for 2024.

A ski slope over a shopping centre. We need one in these parts.

A rain garden is a low-maintenance, wildlife-friendly solution for managing runoff.

It depends where you sit as to whether this is good or bad news, but apparently houses prices will rise 3% in 2024.

I can’t make up my mind whether I like this house or not. It’s perhaps a bit bland overall, but some nice bits.

And future proofing the terraced house for intergenerational living.

Main image credit: “Incline Dynamics: The Kinetic Raceway” A bold, freestanding installation featuring steel and glass in a vibrant, futuristic setting. (DALL-E)