"New York University Global Study Programme"

This was the only Early Decision offer Andy received during his application season, and it has long been his dream school.

New York is a city defined by an exceptionally fast pace: information updates rapidly, change happens at any moment, and people and systems intertwine to form a constantly shifting flow. For Andy, the city's appeal is not merely its surface-level "vibrancy", but rather something closer to an always-on system—one that continuously receives information, responds in real time, and reorganises itself through change.

In such an environment, many things do not have fixed answers; more often than not, one has to make sense of underlying patterns amid ongoing change. In some ways, this also mirrors Andy's own approach to growth.

Throughout his learning and exploration at Wellington, Andy's expression has never been ostentatious, nor has he been in a rush to define himself. Yet when it comes to specific problems, he tends to probe more deeply: How was a particular conclusion derived? What assumptions does it depend on? If those conditions change, would the outcome change as well? This habit has led his approach to learning towards a process of "bottom-up deconstruction". Rather than memorising conclusions, he is more concerned with understanding the structures and logic that underpin them.

He also rarely describes himself in terms of "what kind of student I am", nor does he frame his development in grand narratives. Behind this relatively quiet manner of expression, however, lies a steadily inward-moving process of learning: continuously deconstructing problems, and continuously reconstructing understanding.

To truly understand how this transformation took shape, however, one needs to go further back in time, to an earlier stage of his journey.

Beyond computing, Andy has long maintained another enduring interest: music. He began learning the piano at the age of five, and for many years it formed the most stable part of his life outside the classroom.

However, over time, his understanding of music began to shift.

"For example, if you compose a waltz, it has to be in triple time—it follows that rhythm. If you write a nocturne, it must be elegant. Likewise, jazz is fast and complex," he explains. This perception is not a rejection of music itself, but rather an observation that gradually took shape—music can vary, but such variation often still occurs within certain established frameworks.

By contrast, Andy tends to be restrained in his mode of expression. When music needs to be analysed and deconstructed as a kind of "language", he feels a certain unease. This misalignment did not diminish his interest in music, but it did gradually shift its place in his life—from a primary focus to a long-term personal interest.

At the same time, new "variables" began to emerge.

After joining at Wellington, he was introduced to the Computer Science curriculum here. Compared to the piano training he had devoted years to, with its clear pathway, programming and computational thinking felt more like a new venture. But it was here that he first encountered the holistic structure of 'the computer as a system' – not just using software, but beginning to understand programs, logic, and the mechanisms behind them.

He started writing simple code and tried to understand why programmes produced certain results. In this process, computers revealed a feedback mechanism entirely different from music: not dependent on whether something fits an expected form of expression, but instead on a direct binary judgement—whether the code runs, whether the logic holds, whether the problem is solved.

This was a completely new experience for him.

Gradually, he found himself spending more time on this subject. He began exploring different layers of computer science in sequence: from hardware fundamentals, to programming languages (C++, Python, Java), then to algorithms, and eventually into AI-related content. What had once been scattered interests became reorganised into a path that could extend forward.

During this process, his teacher Diane played a crucial role. Rather than advancing the course at a fixed pace, she adapted the learning path to Andy's needs—accelerating foundational content to free up time for deeper exploration. Through ongoing discussions, she also helped him draw together his dispersed interests into a clearer direction.

"She was really helping me find a line," Andy recalls, "to find a line of my own within this vast world of computing."

Along this line, he began making conscious choices. Networks, security, algorithms, data processing—he explored all these branches, but not all of them held his attention. Some were quickly set aside, while others drew him in for deeper understanding.

Gradually, "AI" began to emerge from among these options.

The reason was not that it was more popular, nor driven by external expectations, but because whenever he encountered related material, he would naturally go one step further: examining the structure of a model more closely, thinking more deeply about its decision-making process, and asking one more question—"why".

Once this path began to take shape, new challenges also became apparent. In the early stages, he needed to fill significant gaps in foundational knowledge, particularly in hardware. "For example, how a hard drive works—I didn't understand any of that at first. I just had to work through it bit by bit," he says.

At the same time, doubts from his family persisted. In their view, whether a child accustomed to a more relaxed pace could adapt to a field like computer science—highly dynamic and requiring sustained, intensive effort—was itself a question.

Andy began to devote more time beyond his regular coursework: systematically practising problems, debugging code, and gradually gaining feedback through competitions and projects. Within a year, he progressed from Bronze to Gold level in USACO (the USA Computing Olympiad). These concrete results gradually made what had once been an abstract "choice" become something visible and tangible.

Looking back, this stage was not about endlessly expanding possibilities, but rather the opposite: the path was gradually narrowing. Starting from an initial interest, entering computing, then exploring its various branches, and finally focusing on AI. With each decision, the direction became a little clearer, until it converged into a defined outcome.

Andy's first systematic engagement with AI began with two consecutive summer programmes.

At Amherst College, he first turned his attention to the question of whether AI is fair. He conducted a study based on medical data, using a model to predict an individual's risk of heart disease, and then analysing whether this process produced different outcomes for different groups. He ultimately wrote a paper examining potential discrimination within medical classification.

This was followed by a summer programme at the University of Pennsylvania, where the pace shifted towards a more technical focus. He systematically studied the fundamentals of deep learning, worked with image-processing models, and took part in training text-based tasks using language models such as BERT. During this stage, his attention centred on how models "learn", how they make decisions, and how different types of data influence their outputs.

These two experiences—one oriented towards questions, the other towards methods—may appear to move in different directions. Yet for Andy, they pointed to the same underlying issue: how exactly does AI arrive at its judgements?

It was also during this process that he began to realise that the answers produced by a model are not necessarily the answers to the question itself.

In his Amherst research, he worked with a set of medical data. At first, everything appeared normal—the model ran successfully, and its accuracy was reasonably high. But when he stopped looking only at the overall results and instead broke the data down by different groups, problems began to emerge. Some groups were more likely to be predicted as "high risk", while others were not—and these differences were almost invisible in the aggregated data.

He attempted to "fix" the issue—changing models, adjusting parameters, retraining—but the results showed no fundamental improvement. Gradually, he realised that the problem might not lie with the model itself.

The issue had existed much earlier.

If the training data already contains certain tendencies, the model merely reproduces them. It does not assess whether they are "fair"; it simply learns existing patterns more efficiently and consistently.

It was at this point that he began to reconsider a more fundamental question: if we do not first define what fairness means, how can a model possibly achieve it?

This issue is particularly evident in medical contexts. For instance, if the data shows that "the older the individual, the higher the probability of illness", a model may further simplify this into "young people are unlikely to fall ill". Statistically, such a conclusion may appear valid, but when applied to real-world decisions, it can be misleading and may even affect outcomes in practice.

In other words, a model being "correct" does not always mean it is "reasonable".

This understanding was further deepened during his time at Penn. Different types of models—whether processing images or interpreting text—ultimately rely on the same thing: data. A model does not truly "understand" the world; it identifies patterns within existing data and uses those patterns to make predictions.

If the data itself is biased, the model will simply express that bias more clearly. This led him to a more defined conclusion: the challenges of AI are not purely technical, but fundamentally about how rules are defined.

Many people ask whether AI can be trusted. In Andy's view, the very question already reveals something important: trust is not something a model inherently possesses—it is something humans must construct. And this construction depends on how we set standards, select data, and interpret results. From this perspective, a more powerful model is not necessarily more neutral; it may instead amplify the assumptions embedded at the outset with greater precision.

After these two summer programmes, Andy's focus within AI began to shift. He continued to study techniques, build models, and participate in projects, but he was no longer concerned solely with whether something could be built. Instead, he repeatedly returned to a single question:

Under what circumstances should this be used?

For him, AI was no longer just a tool, but a system that requires continuous scrutiny. What truly interested him was no longer simply how to make it more powerful, but how to make it more reasonable.

Not every clearly formed path is easily recognised.

More often, university admissions bring together pupils shaped by very different experiences and trajectories, only for their achievements, activities and essays to be placed within the same highly selective framework for comparison. In the end, those who remain are not simply "the most outstanding", but those who are more readily identifiable within that system.

For Andy, this moment of recognition came with New York University.

As a university receiving over 100,000 applications each year, NYU's overall acceptance rate in recent years has fallen into the single digits (approximately 8–10%), and is typically even lower for international applicants. In other words, within any given round, most candidates are not "insufficiently prepared", but rather indistinguishable within a pool of highly capable individuals.

This is why simply demonstrating ability is often not enough. Among large numbers of similarly strong applications, what ultimately makes a difference is something far less replicable: how a person forms their own path, and whether they genuinely understand it.

This is precisely where Andy stands apart.

His experiences were not meticulously arranged around a predetermined goal. From music to computing, and then to AI, he did not follow the most "efficient" route. Instead, through continual experimentation and adjustment, he gradually approached a clearer question: what is it that he truly wants to understand?

For this reason, his application materials did not emphasise what he had "achieved" in isolation. Rather, they presented a continuous evolution—how his focus shifted from the model itself to the judgements behind it, and how, through specific research, he came to realise that "correctness" is not merely a technical matter.

In an environment that values openness and interdisciplinarity, NYU has long placed importance not on how "standardised" a path is, but on whether it is authentic and capable of extension. Compared with highly pre-planned application trajectories, an experience shaped by ongoing refinement of one's own understanding often holds greater potential for sustained development.

In this sense, Andy's offer is not merely entry into a highly competitive university, but entry into an environment that remains genuinely interested in the questions themselves.

Within such an environment, he will share a starting point with others who, in their own ways, have chosen to think differently about the world. Among NYU's alumni are many who have redefined paths within their respective fields—from film director Martin Scorsese, to Lady Gaga, to technology entrepreneur Jack Dorsey. What they share is not a standardised background, but a moment at which they chose to interpret and reshape their fields on their own terms.

Placed back in Andy's context, what he is entering is not a system that merely validates outcomes, but a space that allows paths to continue evolving.

The meaning of an offer, therefore, shifts subtly. It is no longer simply a confirmation of past achievements, but rather a judgement: when faced with more complex problems in the future, does this individual have the capacity to continue advancing their own understanding?

In this sense, being selected is not an endpoint, but the beginning of a more demanding journey.

If we break Andy's development down into a more grounded process, it does not stem from a single decisive choice, nor from one clearly defined turning point. Rather, it is the result of being gradually shaped through the routines of long-term school life.

As a founding pupil, Andy has spent nearly eight years at Wellington College Education (China) - Hangzhou. Much of this change has taken place in details that appear ordinary—perhaps even easy to overlook: classroom discussion, project collaboration, subject choices, and the continual expectation that he should "make his thinking clear".

Before entering an international curriculum, Andy had in fact experienced a more standardised learning pathway. Although not within the public education, the curriculum structure and assessment methods were closer to a traditional model, emphasising uniform pacing, outcome-driven learning, and problem-solving drills. This background provided him with a certain academic foundation, but it also meant that he had to adapt to an entirely different learning logic: shifting from "giving answers" to "explaining processes", and from "completing questions" to "expressing viewpoints".

One of the most immediate changes was in his ability to express himself.

When he first arrived at Wellington, Andy did not have a strong advantage in verbal expression. He could carry out logical reasoning and understand complex ideas, but when required to articulate his thoughts clearly in English, his structure would often jump, sentences might remain incomplete, and his ideas would sometimes move faster than his language. This was not resolved through any single intervention; rather, it was gradually corrected through sustained classroom interaction.

In English lessons, academic writing sessions, and daily discussions, teachers consistently asked him to reorganise fragmented thoughts into coherent arguments. At times, they would even refine his sentences line by line, helping him recognise that thinking and expression are not the same thing. Over time, he developed a new habit: before voicing an idea, he would first consider whether it had been properly structured.

This change became particularly evident in later stages. Whether chairing Model United Nations sessions or delivering classroom presentations, he was required to organise information within limited time, manage pacing, and respond clearly to others' questions. Expression was no longer simply about "speaking out", but had become a skill that could be trained and refined.

Another, perhaps more significant, change came from the collaborative environment.

Andy's initial inclination was to work independently. He preferred solving problems on his own rather than relying on others. However, through repeated collaborative experiences, he gradually realised that this approach was inefficient for complex tasks and could even limit the overall quality of outcomes. Over time, he adjusted his approach—from "completing his own part", to "understanding how the whole project operates", and then to reconsidering "what role he should take within it".

This shift was not theoretical, but formed through repeated practice: some team members drove progress, others managed communication, others refined details—and he had to learn to move between these roles.

Beyond the classroom and projects, boarding life itself became a form of implicit training. In a highly diverse environment, differences arising from varied cultural backgrounds are immediate: communication styles, habits of expression, and interpretations of rules are not always aligned. Often, the issue is not one of right or wrong, but of how something is understood.

Within such an environment, he gradually learned to adjust his own way of communicating, while becoming more sensitive to how others receive information. This ability is not overt, yet it accumulates over time and becomes a stable form of adaptability.

If we consider these changes together, they do not point towards a single outcome. Instead, they form a more fundamental structure of capability: the ability to express, to collaborate, and to continuously adjust oneself within uncertain environments. These abilities may not directly appear on transcripts or in offer letters, but over a longer time horizon, they determine whether someone can enter more complex systems—and continue progressing within them.

In this sense, what Wellington provides is not merely a curriculum or a defined pathway, but an environment that consistently requires pupils to "make themselves understood" and "carry things through". It places them repeatedly in real contexts of expression, collaboration, and public discourse, allowing abilities not just to be trained, but to stabilise through continual use.

It is within such an environment that Andy completed a shift—from a learner inclined towards independent thinking, to someone able to collaborate within structures, articulate ideas, and continuously adjust his position. These changes may appear gradual, but once accumulated, they form a remarkably stable foundation.

Listening to Andy reflect on his years at school, many of his experiences are described in terms that are interesting, enjoyable, and broadly positive. He does not tend to explain meaning through lists, nor does he habitually break experiences down into neatly attributed outcomes. What seems more important to him is an overall sense: of being carried forward by experience, while also remaining willing to continue moving forward.

From first understanding "rules and structure" through music, to realising in computing that "rules are not fixed, but can be constructed", and then gradually entering into AI-related study and reflection, his way of understanding has been continually evolving. Yet this evolution is not the result of any single decision, but a direction that has gradually taken shape through ongoing experimentation and adjustment.

If, in the past, his understanding of the world and of computing was largely bounded by school and curriculum, that boundary is now steadily expanding. New environments, new questions, and new systems continue to enter his field of vision. And within this process, what becomes truly clear may not be how far he has already gone, but that he remains willing to keep going.

The path is still unfolding, and the answers are still ahead.