Why astronomy and AI belong together
Julia chose astronomy not in spite of its practicality, but because of it. "Astronomy and cosmology is a field with a lot of big data components. We have enormous amounts of data from telescopes," she explains. "Anything you learn that you can apply to astronomy is easily transferable to any other data science field."
It's a smart hedge: pursue the thing you love, while building skills that transfer anywhere. Right now, the field sits at the exact intersection of where AI is making its most dramatic impact: vast structured datasets, pattern recognition at scale, and the kind of inference workloads where models like Claude and GPT are genuinely indispensable.
How researchers at Cambridge actually use AI
Here's the part that surprises most people outside academia: at Cambridge's research level, the shift to AI assisted work has already happened. It's not hypothetical. Among the younger PhD cohort Julia works with, the change is stark.
It's been roughly six months since any of us wrote a line of code on our own. We delegate the raw writing to LLMs. The architecture, the problem framing, the judgment? That's still ours.
Julia, Cambridge PhD Researcher
The fundamental skills question
This is where the conversation gets genuinely complicated. If junior researchers and professionals are no longer writing code from scratch, how do they build the foundational intuition that makes AI assistance actually valuable?
Julia's answer is careful, and probably not what most people expect to hear from someone her age.
If you already know how to solve a problem and you ask an LLM to help, you can guide it so much better. You know where it might go wrong. It smells wrong to you. If you don't understand the core concepts, you're unable to tell what's right from what's wrong.
Julia
The implication for education is significant. AI is making in person, unassisted examination more important, not less. As Julia observes from her supervising role: anything you can take home, you can now pass with an LLM. The only remaining way to verify genuine understanding is the old fashioned written exam, just you, a white sheet of paper, and what you actually know.
At Cambridge, this is already reshaping how assessments are designed. The challenge is ensuring the next generation doesn't skip the hard miles entirely, because the intuition you build from struggling through problems yourself is exactly what makes you a good AI operator later.
What Gen Z understands about AI that older professionals don't
Julia's answer here is less about technical skill and more about something harder to teach: comfort with constant change.
"Maybe older, more experienced professionals have been working the same way for long years and they're really good at it. That's why they're where they are," she says. "But during their period of professional growth, there were fewer changes. For us, since high school, the speed of change has been so fast that we're really used to adapting to anything new."
What makes the next generation different with AI
- Comfortable switching tools weekly, no emotional attachment to a single model
- Treat AI as a collaborative layer, not a shortcut or a threat
- Discuss what's working across peer networks regularly, crowdsourced benchmarking
- Apply the same critical thinking to AI outputs as to any other data source
- Never buy annual subscriptions, flexibility is the whole point
This isn't just adaptability as a personality trait. It's been trained into them by the pace of the environment they grew up in. Knowing what to pay attention to in a fast moving information landscape, and knowing what to ignore, turns out to be a genuine professional skill, one that didn't matter much before AI, and now matters enormously.
On the future of junior careers
Julia's take on the job market is measured, and probably more honest than most think pieces on the topic. Her immediate circle is largely doing PhDs, so the pressure isn't acute yet. But she's clear eyed about what's happening elsewhere.
"A lot of jobs that used to have meaning will be automated," she says. "Unless people learn to use AI to improve their work and sit on top of it. Because some things that used to be their task are just going to be replaced."
Andrea's observation from Nova reinforces this: the gap between the top performers (those who have genuinely integrated AI into how they work) and everyone else is widening faster than any previous technology shift. It's not that junior jobs are disappearing uniformly. It's that the productivity differential between the top 10% and the average is bigger than it's ever been.
AI won't replace recruiters, but it will expose the average ones. The same is true for every field.
Nova Signal recurring theme
On paper inflation and the quality problem in research
One concern Julia raises that rarely gets airtime: AI isn't just accelerating good research. It's also inflating the quantity of output without necessarily improving the quality, and in academia, where career progression is still heavily tied to publication count, that's a real problem.
"Academia has always had a tendency to focus on quantity over quality," she says. "LLMs have amplified this. You're able to publish so much more, but this doesn't necessarily mean you improve on quality."
The research she references, a paper by Lani Perez, points to a growing literature on this. The worry isn't that AI makes bad research. It's that it makes it very easy to produce a lot of output that looks like research, crowds out the signal with noise, and creates the appearance of progress without the substance.
The same pattern shows up in recruiting, in marketing, in content. Everywhere AI dramatically lowers the cost of production, you get more output and a harder job filtering for what actually matters.
Key takeaways from this episode
- The AI shift in research has already happened, younger Cambridge researchers haven't written raw code independently in months
- The right AI workflow is layered: design with the strongest model, implement with a faster one, cross check across both
- Fundamentals still matter, AI is an accelerator on top of real knowledge, not a substitute for it
- In person, unassisted exams are becoming more important in education, not less
- Gen Z's advantage isn't knowing more about AI, it's being comfortable constantly updating how they work
- Annual subscriptions are a trap, the best model changes week to week
- The gap between top performers and everyone else is larger than any previous technology shift has created
- AI raises output quantity everywhere, the new challenge is filtering for quality
Join the network that finds people like Julia before the world does.
Nova is a curated community of world's most exceptional talent, selected based on merit. If you're building something remarkable or want to be around those who are, this is where to start.
Apply for the Nova membership →
