The expert in your pocket
Finding practical uses for AI while the world ignores o3
The reaction from the wider world to o3 has been a collective shrug. Much of the mainstream media doesn’t seem to have noticed yet. Take BBC News - there isn’t a mention of o3 on their website. OpenAI have effectively said “we’ve invented a pill which can more than double human IQ but it might have some side-effects like radically changing whole areas of science and engineering”. And the reaction of the world? Let’s eat turkey and focus on dinosaur footprints.
Humans are weird.
While we wait for the rest of the world to catch up with o3 I’ve been exploring other uses of AI over the Christmas break. First up was getting Claude to help me with some home improvements…
Getting warmer
Our house is a typical 1930s bungalow. It’s been modified over the years - an extension here, new windows there, a replacement door… Some rooms are warm, but the front rooms face north and have thirty year old double glazing. They get quite chilly in the winter. And I spend my days sitting next to one of them. So I’ve been considering upgrading the windows to triple glazing.
Normally I’d sit down with Excel and do some rough sums. But this time I asked Claude…
Claude did some calcs and estimated the annual heating required as being ~15000kWh broken down as:
Because I’m, err, me, I’ve got a spreadsheet tracking gas usage. And last year our heating cost was 14574kWh. Pretty neat.
Claude assumed modern construction that complies with current building regs. And while our house is nearly 100 years old, we demolished large chunks of it 10 years ago and rebuilt to modern standards - and upgraded the rest. So the assumptions are reasonable.
Then I asked Claude to compare the spreadsheet with its estimate. Claude thought a lot. It wrote Javascript to parse the spreadsheet and then tried. And tried again. And again. But it struggled. When it did produce a number it decided the total was 20000kWh - and I’m not entirely sure why. The spreadsheet is quite complex, but ultimately it’s just a big table with a lot of rows.
Then I asked about savings. Claude reckoned £175 per year at 7p per kWh (I currently pay 6.25p per kWh). With a payback period of ~30 years. On the more positive side, Claude reckoned the temperature next to the window would rise from 13C to 19C when it was 0C outside - so I would be less chilly when I was working.
Which is exactly what I’ve found - I upgraded the windows in mid December :).
Debugging power supplies
In my free time I like repairing old electronics. Over Christmas I recapped the power supply for a 30 year old PC. And it all worked fine until… …I accidentally shorted the -12V rail to ground. Sigh.
But Gemini to the rescue. I took a couple of pictures and, perhaps optimistically, asked Gemini to reverse engineer the circuit schematic…
Unsurprisingly Gemini refused - reasonably pointing out that it couldn’t see all the components. However it did work out that this was a switched mode PSU. Cool. And it tried to identify the main components - some of which it got right (e.g. transformer, reservoir caps) and some wrong (the switch mode controller isn’t a bridge rectifier).
Where it became useful was in explaining the operation of this design. Close-up photos enabled it to work out the part numbers of the main ICs - and it told me how they were likely to be configured. Very handy - no more looking up part numbers in data sheets. Then I explained how I’d shorted the output and Gemini was only to happy to offer debugging suggestions.
It suggested I check the TL494 (the main switch mode controller) was powered by checking for a voltage on pin 12:
0V on pin 12. Aha!
I explained the connections between pin 12 and the rest of the circuit to Gemini. And together we worked our way back to the primary side and discovered one of the push-pull switching transistors had died.
Gemini wasn’t able to immediately solve the problem. But it was an excellent assistant. It knows the pinouts of common ICs. It knows all the various different switch-mode PSU designs. It knows the likely way the primary and feedback transformers are wound. I appreciated the help. It no doubt helps that I’ve spent many hours repairing switch-mode PSUs. But it genuinely saved me time - and the partnership made the process more fun too.
Now imagine Gemini with real-time video and audio. Gemini can see everything you are looking at. The PCB, the oscilloscope, the voltages. And you can discuss trouble shooting with it. That’s going to be a step change. And it’s not just electronics repair that will be impacted… …think of all the situations which benefit from an expert-on-your-shoulder. There are thousands.
BTW if you’re going to play with switched mode PSUs be careful. There is ~350V across the two main reservoir caps. You really don’t want to touch that…
So what?
From estimating home heating efficiency to debugging electronics, AI is becoming a practical tool for everyday problem-solving. While it's not perfect, it's already an invaluable assistant-on-your-shoulder. As models like o3 emerge and video/audio interfaces improve, the partnership between human expertise and AI assistance will only grow stronger. The future isn't about AI replacing human knowledge - it's about finding creative ways to combine both. And the only way to discover those opportunities is to keep experimenting, one project at a time.