Notes on learning, safety, and building an AI a parent can trust.
28 articles
A chatbot that answers instantly feels helpful — and quietly makes learners worse. Inside every OpenKids coach is a teaching contract that no request can talk its way around.
ReadThe guard at the gate, the form teacher who reads the question first, the teacher, a check before anything is said aloud, and a note home. Our safety pipeline, told in the language of a school day.
ReadNo streaks, no daily-login gifts, no fear of losing progress. Here is why the most common engagement tricks are banned inside OpenKids.
ReadTrust is not a slogan; it is an interface. A precise tour of the parent dashboard, the weekly report, safety alerts — and our data-minimisation red lines.
ReadOne generic assistant cannot teach P5 fractions and JC calculus well. Thirty specialist coaches, organised like a school, can.
ReadThe fastest way to make a child dependent on AI is to answer them. The slower, harder way is the one that actually teaches.
ReadWhen the user is a child, safety cannot be a setting you switch on. It has to be the thing every other decision is built on top of.
ReadThe point of putting AI in front of a learner was never the intelligence of the machine. It was the understanding left behind in the child.
ReadYes for practice and understanding, no for copying answers — and the tool's design, not the child's willpower, decides which one happens. A straight answer for parents.
ReadHuman tutors read the room and carry relationships; AI never tires, never judges, never sleeps. An honest comparison from people who build the latter — and say both.
ReadChatGPT is a remarkable tool built for adults. Whether it suits your child is a question of fit, not villainy — answered honestly by people who build children's AI.
ReadTeaching by asking, not telling: what Socratic tutoring is, why it survived twenty-four centuries, and why a machine that never tires is unusually good at it.
ReadThe same technology that can hollow out a child's thinking can strengthen it. The difference is not the model — it is where the design chooses to put the effort.
ReadMinutes measure exposure, not learning. A saner way to judge screen time: ask what the child is doing on the screen — and whether the session was built to end.
ReadPrecise location, contacts, ad identifiers, behavioural profiles for advertising — the data a children's education app should never collect, and how to check.
ReadWhen a child tells an AI something worrying, the honest measure of the system is simple: does it make sure a human shows up? Here is exactly what ours does.
ReadStreaks punish sick days and manufacture anxiety. Every badge in OpenKids marks a mastery event — a first proof, a fixed misconception — never hours or days.
ReadFive questions that work on any AI product for children: can you see the conversations, does it teach or just answer, what does it collect, what happens in a crisis?
ReadBar models look slow next to algebra. They are not a detour — they are training in making the unknown visible, and that is exactly what PSLE word problems test.
ReadMemorised phrases make every composition sound the same. What improves primary school writing is detail you can see, a planning ritual, and rewriting well.
ReadChinese language sense will not come from spelling lists. It grows from comprehensible input plus retrieval — and ten minutes a day beats two hours on Sunday.
ReadThe mistake book is the highest-yield revision tool we know: retrieval practice aimed exactly where your knowledge breaks. Here is how to keep one properly.
ReadExam anxiety in children is information, not weakness — and rehearsal beats reassurance. What helps the night before, the morning of, and in what you say.
ReadA chatbot is one voice that adapts to you. A school is many voices with standards you adapt to. That difference is the whole reason OpenKids AI exists.
ReadSingapore households spent about S$1.4 billion a year on private tuition at the last count. The real cost is not the money — it is who gets left out.
ReadA persona with one job can hold standards a do-everything assistant cannot. Inside the design decision that shapes everything else about OpenKids AI.
ReadOpenKids AI coaches are not translated — each is written natively in English and in Chinese. Why that distinction decides whether an explanation lands.
ReadEvery ed-tech product in Singapore claims alignment. Here is what the word means inside OpenKids AI, what it does not mean, and the question to ask any vendor.
Read