Q&A

Start with ChatGPT or Claude — they're the most capable general-purpose tools right now. Don't overthink it. The specific tool matters less than learning how to work with AI effectively. Once you understand the principles, switching tools is easy.

It depends on the data and the tool. Commercial tools like ChatGPT and Claude have different data policies — read them. For sensitive data (patient information, unpublished research, proprietary data), consider local LLM solutions or enterprise versions with data processing agreements. When in doubt, anonymize first.

One focused day can change how you work. But the real learning happens through daily use. Start small: pick one task you do regularly and try doing it with AI assistance for two weeks. That's where the insights come from.

You can use AI to help you write — drafting, editing, restructuring, improving clarity. But the ideas, arguments and intellectual work must be yours. Think of AI as a writing assistant, not a ghostwriter. Most journals now require disclosure of AI use, and reviewers can often tell when text is AI-generated. Use it to write better, not to write less.

Never let AI write from scratch. Start with your own draft — even a rough one — and use AI to improve it. Provide examples of your previous writing. Edit everything the AI produces. Your voice comes from your choices, your examples, your edits. AI is a tool for refinement, not replacement.

No — as long as the intellectual content is yours. Funders care about the quality of your ideas and your ability to execute. Using AI to improve clarity, structure or language is no different from using a writing center or a professional editor. Just be transparent about it if asked.

You verify it. AI can be confidently wrong — this is called hallucination. Never trust AI output for facts, citations, or technical claims without checking. Use AI for structure, language and ideas. Use your expertise and external sources for verification. The human in the loop is not optional.

They're real and they're dangerous if you don't expect them. AI models generate plausible text, not verified truth. They will invent citations, fabricate statistics, and make up facts with complete confidence. This is why we teach verification workflows as part of every program. Assume everything needs checking.

No. AI can help you prepare for peer review — catching errors, improving clarity, identifying weaknesses. But peer review is about expert judgment, domain knowledge and scientific discourse. AI lacks the context, the expertise and the accountability that peer review requires.

Prompting is about the words you use to ask. Context engineering is about the information you provide before you ask. Most people focus on prompts — the right magic words. But the quality of AI output depends far more on the context you give it: background information, examples, constraints, format requirements. Get the context right and prompts become almost trivial.

Structure your inputs. Provide clear context every time. Use templates for recurring tasks. Save prompts that work well. The more you standardize your inputs, the more consistent your outputs become. Randomness in results usually comes from vague or inconsistent instructions.

Yes — carefully. AI is excellent for learning programming basics, explaining code, and helping you write simple scripts. But don't run code you don't understand. Start with small, low-stakes tasks. Use AI to learn, not just to produce. Understanding what the code does is more important than getting it to run.

AI will replace tasks, not researchers. The researchers who learn to work effectively with AI will have an advantage over those who don't. The core of research — asking good questions, designing experiments, interpreting results, building knowledge — remains human. AI changes how we do research, not whether we do it.

It's coming and it matters. The EU AI Act classifies AI systems by risk level and imposes requirements accordingly. Most research and clinical AI use will fall under limited or minimal risk categories, but some applications — especially in healthcare — may face stricter requirements. Start understanding it now. We work with specialists on this.

Focus on principles, not tools. Tools change every few months. But the fundamentals — problem decomposition, context engineering, quality verification, human oversight — these stay relevant. Learn the methodology and you can adapt to any tool. Chase features and you'll always be behind.