Working with AI

In my daily work

I work with AI as I would with a capable research assistant: it drafts and suggests; I check, decide and take responsibility. I label its part wherever it applies. An AI summary tells me whether a paper is relevant; it never means I have understood it.

• I have started to build LSAP, the Legacy Survey of astro-ph, a literature management system named after the Rubin Observatory's Legacy Survey of Space and Time. Its deliverables will be a reading list, drawn from the daily scores of MapHub and the references of papers I know, and one file of notes per paper, drafted by AI from the paper's arXiv source, with every entry marked as sure or unsure; I will read every one.
• When I write a paper with AI, I give the message, the idea for the figures and every judgment about what the work is for; AI drafts, searches the literature and writes the code that computes the numbers, which I check. Before my coauthors see a draft, reviewers simulated by AI, from different subfields, read it, first carefully and then as busy readers skimming for two minutes. Placeholders mark what is mine to supply and what is simulated, so that nothing unfinished goes out. No citation goes in unchecked: I check every reference in ADS.
• This website is written with AI assistance (Claude) and reviewed by me, as every page's footer says.

Teaching with AI

For college teachers trained in the traditional way, generative AI is both a challenge and an opportunity. It can help teachers design courses and activities, join brainstorming and discussion, and give students personalized support and feedback. For students, AI literacy is now essential, and they need guidance on using AI effectively and responsibly. A goal I find achievable: a successful student will be able to do something, with or without AI, that someone else cannot easily become able to do, with or without AI. In short, students should give better answers than AI, not just answers with it, and they should learn to check AI results, for example with null tests.

In 2025 I wrote a handout for fellow graduate students and college teachers, A College Teacher's Survival Guide in the AI Era, based on José Antonio Bowen and C. Edward Watson's Teaching with AI (2024). The tools have changed quickly since then, but its main points still hold for me:

• People, not tools, are accountable for results. Students are responsible for what they submit.
• No assignment is truly AI-proof, and no detector is perfect. Cheating often means students do not see the value of doing the work themselves, so the incentives need to be there.
• Neither banning AI nor allowing it without limits works best. Discuss AI policies with students early, starting with why.
• Used well, AI can be a patient tutor, help personalize examples, and take part in class discussion, for instance as devil's advocate.

How I teach more generally is on Teaching Philosophy.

Acknowledging AI

In research, AI plays two main roles: something like a search engine, and a writing assistant. The first does not need to be mentioned; the second should be acknowledged. Every page of this website says that it was written with AI assistance and reviewed by me.

Applying for jobs with AI

I used an AI chatbot a lot when applying for postdocs, and corrected it often. It invented facts about me, got eligibility rules and page limits wrong, rated almost every position an "excellent match," and lost track in long sessions. What worked:

• Give it only verified sources about yourself (CV, abstracts, exact statements of your contributions), and tell it not to make things up.
• Ask it to explain any phrase before you keep it. You should be able to support every sentence you send.
• Ask for small edits that keep your own words, one task at a time.
• Use it for reading, summaries and structure. Check facts, fit, names and rules yourself.

More from that search is on Career Development.