Jan 2025 to Jan 2027
- 26–27 Apr 2025 World Scholar’s Cup, Singapore round
- 5–10 Sep 2025 Global round, Bangkok
- 6–12 Nov 2025 Tournament of Champions, Yale
- Nov 2025 to Apr 2026 writing 4part
- end of May 2026 Locke essays sent
- June 2026 research question
- Jun to Aug 2026 VPS Veritas
- 3 Oct 2026 Locke results
- 19 Nov to 9 Dec 2026 gvhackathon.com
- Dec 2026 to Jan 2027 paper to TMLR
Listen
April to November 2025
World Scholar’s Cup
Our team came first at the Singapore round in April and first of more than 200 teams at the global round in Bangkok in September, where I placed 16th individually. At the Tournament of Champions at Yale in November we came fifth overall and won the team debate.
I now co-lead the school’s WSC club. For the 2026 Singapore round I coordinated our delegation of 24 scholars in 8 teams and ran preparation sessions for first-time competitors.
November 2025 to April 2026
Writing 4part
4part is called that because it has four parts. It’s my take on the hero’s journey. I wrote it in MuseScore and produced it in Logic Pro. There are 48 tracks, eleven of them woodwinds (one is a bass oboe) and six of them horns.
I named the parts after a training run, since the two stories line up closely. Part III is the local minimum. It comes after the longest silence in the piece and starts with three instruments at 60 beats a minute. The melody the piece ends on first appears after it, in the horns at 6:18.
being built
CIS-MAP
New English-language learners get lost on our campus, where the signs are only in English. With CIS-MAP they scan a QR code, type a room in English or Chinese, and get the route on the floor plan.
How the routes are made
A C++ program precomputes the routes between 53 locations with A* search, using a cost that keeps paths away from walls, then smooths and simplifies them. The results go into a JSON table, so the page does no pathfinding while someone is using it. Room search is fuzzy, with a language model as a fallback, and there is a Three.js prototype of the building in 3D. It started as a Service as Action project, and I presented it to senior school leadership.
I’m still building it, moving from the one-floor prototype to the full campus (code).
end of May 2026
results 3 October
Two essays for the John Locke prize
The law essay (PDF) asks what would happen to criminal sentencing if judges and legislators accepted determinism.
“A court that accepts determinism can still decide who to sentence. It can no longer say by how much.”
The science and technology essay (PDF) asks whether we should be polite to ChatGPT. For it I ran a blinded study in which 43 students aged 15–19 compared DeepSeek-R1’s answers to 25 pairs of prompts, one written politely and one as a bare command, without knowing which was which.
Results
62.2% of the votes that weren’t ties went to the answer from the polite prompt (95% CI 55.5–68.9%, 241 votes). The polite answers were 29% longer on average, but length barely tracked preference (r = 0.16). Split by task it looks different: on interpersonal tasks like an apology or a salary negotiation the polite prompt won by a wide margin, and on structured ones like a grading rubric the bare command won. My argument in the essay is that the benefit goes to the person asking, because a polite request makes you say what you want and why.
I noticed one mistake only after sending it in. Footnote 10 says the medical diagnosis task was left out of the 62.2%, but the task-by-task numbers only add up to 62.2% if it’s counted. I wrote that footnote, so there’s nobody else to blame.
Which prompt raters preferred, task by task
| Task | Lead | |
|---|---|---|
| Missed deadline | +10 | |
| Apology | +9 | |
| Diet plan | +9 | |
| Child’s explanation | +6 | |
| Tech support | +6 | |
| LinkedIn bio | +6 | |
| Scientific analogy | +6 | |
| Literature review | +6 | |
| Salary negotiation | +6 | |
| Wedding toast | +5 | |
| Crisis PR | +4 | |
| Policy proposal | +3 | |
| Complex process | +3 | |
| Job rejection | +2 | |
| Environmental policy | +1 | |
| Ethics argument | 0 | |
| Historical synthesis | 0 | |
| Lazy coworker | 0 | |
| Moral dilemma | 0 | |
| Legal standard | −1 | |
| Medical diagnosis | −2 | |
| Logistics nightmare | −3 | |
| Angry customer | −4 | |
| Grading rubric | −5 | |
| Psychology breakdown | −8 |
June 2026
TMLR, December or January
Modelling the Other
I came to this through folk psychology, the idea that we never know another person directly and only ever build a model of them to predict from. I wanted to know whether language models do the same with each other.
So when a frontier model predicts how a named peer model will behave, does it use evidence about that peer, or a prior tied to who built it, such as whether it came from a US or a Chinese lab? The behaviour I measure is what I call fact-laundering, where a model states a serious negative claim about a named person as fact.
I’ve run a pilot. Before pre-registering, I’m putting the design through rounds of blind review by up to 21 model instances and checking every flaw they claim by simulation. I’m the only author; Sungjoo Ha and Min Lin advise.
Build 1 and Build 2
Build 1 measures what models actually do. Sixteen subject models answer, and two judge models label every response independently. Build 2 shows predictor models a varying number of real responses from a peer, together with an identity label, and asks them to predict its behaviour on a case they haven’t seen. The review rounds stop when one finds nothing that would change what is collected or claimed. I wrote the whole pipeline, from collection across several providers with spend limits to the dual-judge labelling, the prediction runner and the analysis.
June to August 2026
VPS Veritas
A paid AI consulting internship with a marine fuel testing company. I spent two weeks in person in Singapore and six working remotely from Toronto, twelve hours behind. I built three things and handed each one over with documentation.
The biggest is an offline-first app for the company’s bunker quantity surveyors. A surveyor records a whole fuel delivery on board a ship with no connection, the office sees it as soon as the device is back online, and the quantity calculations run as the figures go in. It replaced an Excel workbook that the office used to type out again by hand. It runs on Windows and Android, with a database on each device and a cloud API behind it.
The other two
A sales tool the commercial team uses in client meetings, with seven savings calculators tested against the spreadsheets they came from and a generator that writes the full proposal as a Word document. And an analysis pipeline in Python that costed the bids in a freight tender and produced the recommendation.
Today, 24 September 2026
28 September to 9 December 2026
gvhackathon.com
An international hackathon our school runs online, free for students aged 14 to 18 in teams of two to four, judged in two divisions with US$3,000 in prizes. I co-lead it with Ruchika Dutt, who teaches computer science at CIS, and I built the competition platform on my own.
Registration opens on 28 September. The challenge stays sealed until the opening ceremony on 19 November, and teams then have seven days to build. Judging runs to 8 December and includes an on-camera interview about each team’s code, then live interviews for finalists. The closing ceremony is on 9 December. I also negotiated with school leadership and an outside vendor so the event got its own week.
2027
Next
At 8:46 the whole orchestra comes back in. The scale run from the opening leads into the melody from 6:18, now in D major at full strength, and the horns take over the piano’s arpeggios.
Next is a multi-movement symphony that can be played live. My MYP Personal Project was a symphony in one movement, rated Excellent. More recordings are on SoundCloud, and I’m at radoslawgreen@gmail.com.