A techie way to write prompts, a smarter way to search with AI, a tutor that few pupils used and a robot boat that crossed the Atlantic.
I took a short and lovely spring holiday, ten days that took in the end of September through to the first few days of October.
Then I got back to my desk, with some reluctance, and did some triage. Quick tasks for clients, collegial check-ins, the making of lists.
Also – I gradually got to grips with my email. No surprise that there were about 100 missives in the folder called “AI general”. I worked my way through them, rather wishing I wasn’t doing it.
From AI agents behaving badly to US presidents supposedly taking action, and from a tech company talking to religious leaders about AI consciousness to the release of a whole bunch of new models and bots, it was terrifying and overwhelming.
But amid the Really Big Things going on, there were a couple of smaller nuggets of information, along with some generative AI-related tips and tricks. These I could get my head around – and I hope you can too.
A new way to talk to generative AI
Andrej Karpathy, the OpenAI founding member who defected to Anthropic, has made various suggestions about ways to get outputs from generative AI on the platform formerly known as Twitter.
One of the suggestions intrigues me. He recommends asking an LLM to explain something in ASD-STE100. It turns out that ASD-STE100 is an international standard to guide the writing of technical documentation. It’s a product of the Aerospace, Security and Defence Industries Association of Europe, Brussels, Belgium.
I did not know that this standard existed! But once I did, I had to try out his tip.
So I asked Claude (model Sonnet 5.5 Medium, free account), well-known for its verbosity, to “explain the concept of gravity in one paragraph in simple English”. It cogitated (Claude has a variety of ways of describing its inner workings). I got this:
Gravity is the invisible force that pulls objects towards each other. Anything with mass, from a tiny pebble to an enormous planet, has gravity, and the more mass something has, the stronger its pull. This is why the Earth, which is very massive, pulls everything on its surface towards its centre, keeping our feet on the ground and making a dropped cup fall rather than float away. The same force keeps the Moon circling the Earth and the Earth circling the Sun. We do not feel the pull of small objects around us because their gravity is far too weak, but with something as huge as a planet, the effect is powerful and constant.
I then did a new chat, and asked it this: “Explain the concept of gravity in one paragraph using the ASD-STE100 standard”. It crystallised and honed and came up with this:
Gravity is a force that pulls objects toward each other. Every object that has mass causes this force. An object with more mass causes a stronger pull. The force becomes weaker when the distance between the objects becomes larger. The Earth has a very large mass, so it pulls objects toward its center. This is the reason that a dropped object falls to the ground. Gravity also keeps the Moon in orbit around the Earth and the Earth in orbit around the Sun. Isaac Newton described gravity as a force between two masses. Albert Einstein later showed that mass causes space and time to curve, and this curve makes objects move toward each other.
The language in the second is cleaner, and has more detail on the background – but I think the first one is better. I’ll be using the standard in prompts in future though, just to see what happens.
The article where I first read about this notes that STE isn’t meant for general writing, and quotes a commenter on the Karpathy post as saying it doesn’t work for marketing. But wouldn’t it be nice if marketing content could be held to the same standard as technical documentation?
How to search with AI
This will change how you search with AI – This is a long piece by Grammar Girl Mignon Fogarty. Inspired by an Instagram Reel, she went down a rabbit hole of note, trying to figure out how to improve the results one gets when using generative AI to search (and especially to get it to give you accurate results about facts that might have changed recently).
The post is long and detailed, based on one query about a women’s swimming record. One conclusion is particularly useful: don’t ask LLMs leading questions (“Is XXX good treatment for lower back pain?”) or for a yes/no answer (“Is it correct that X holds the record?”). In both of those cases, you are less likely to get an accurate answer.
Generative AI in education: it’s complicated
This headline caught my eye: Scientists Gave 12-Year-Olds Access to an AI Tutor and You Probably Already Know What They Did With It
The article says (in brief) that an AI tutor was given to pupils in 18 schools in Tennessee, and that the children mostly chose not to use the tutor (called Khanmigo), or asked it to just give the answers. There were also off-topic questions: the history of pizza, for instance.
Amusing as that is, there’s nuance.
The 12-year-olds in question were in remedial maths classes and were not necessarily all 12. The paper says they were in Grades 6–8, which corresponds roughly to ages 11 to 14 in the United States.
And there were other factors at play. As the actual research paper concludes:
Conversation with the [AI] tutor… required student initiative,and student initiative was rare: nearly all students tried Khanmigo, but the median student did not message it on most practice days, in most exercise sessions, or in most sessions containing a mistake. The pattern reflects a familiar finding in the economics of education: interventions that rely on student initiative reach fewest of the students who would benefit most… Struggling students are the least likely to seek help unprompted, and the barriers to doing so bind when students face immediate costs and uncertain long-term benefits, especially when actions being considered are incremental, like considering one mistake on one math problem. The limited dialogue we document is thus less a defect of the AI tutor than a lack of interest in attention and learning. (my bold)
Put that together with this reported interaction between a student (S) and Khanmigo (K):
K: What do you notice about the hour hands on the clocks?
S: idk [I don’t know]
K: No worries. Look at the short hand. What number is it pointing to?
S: whats the answer
K: What do you notice about the rectangle’s rows and columns?
S: just give me it I already got it wrong so theres no point doing it again
That last line – my heart goes out to that pupil!
My takeaway? We get an insight into the complexities of the classroom, and (yet again) a lesson in not taking that first headline at face value. (There’s a much better version of the story here: Students rarely engaged with Khan Academy’s AI-powered tutor Khanmigo, study finds – Chalkbeat )
And then there’s AI in the workplace
This should come as no surprise to anyone who’s ever had a toxic boss.
An Inc report says that Glean’s 2026 Work AI Index surveyed 6,000 full-time digital workers across the U.S., U.K., and Australia. It found that 87 percent now use AI at work – and 61 percent say AI helps them more with their day-to-day work than their manager does.
A sailing robot
This is enchanting. A robotic boat has crossed the Atlantic Ocean unaided. The boat, which sadly doesn’t seem to have a name apart from “C-Star PC13”, covered a total of 2,926 nautical miles (5,400km), sailing from Gran Canaria on May 7, 2026 and crossing the finishing line east of Barbados on September 13.
Another little boat made by the same company (Oshen) was the first uncrewed surface vehicle to capture and transmit data from inside a Category 5 hurricane in September 2025.
A little digging got me more detail: C-Stars are 1.2 metres long and weigh around 40 kilos. They have solar panels which power navigation, communications and sensors, while a sail provides propulsion.
Final word
Sometimes the only way to deal with the tsunami of information we all face every day is to look at the smaller things. like little boats that tackle big oceans.
Main picture: Adhitya Ginanjar, Unsplash
Previous Sensible Guide articles
How to write a good prompt for AI – A common complaint about AI is that it just doesn’t give good answers. That’s because you are asking questions. Instead, here’s my guide to how to write a good prompt.
A parent’s guide to Gen AI: 11 ways to help your offspring stay safe and learn stuff – I take my experience as the mother of a gamer to bring you a parent’s guide to AI.
How do you actually fact-check something that Gen AI told you? – It’s now a known thing that tools like ChatGPT “hallucinate” (ie make things up). But how do you fact-check the stuff they say?
If any of this is relevant to your work…
You might find one of these useful:
