Expert insights with Ian Mulvany
From fine tuning to ethics: navigating AI’s role in research
In the first episode of Chats, a new series from Wiley, our chief technology officer Ian Mulvany joins Ray Abruzzi, senior director of AI product management at Wiley, to discuss the practical challenges and opportunities of AI in research and publishing.
They explore why waiting is not an option when it comes to AI adoption, and what it means for developers, publishers, researchers, and institutions. From finetuning models and prompt engineering to assessing performance risks and debunking misconceptions about hallucinations, the conversation offers practical insights for anyone navigating this fast moving space.
Highlights from the conversation
- Fine tuning AI models, prompt engineering, and deployment strategies in real research settings
-
How to evaluate AI tools and build transparent publisher–developer partnerships
-
Clearing up misconceptions about hallucinations and model limitations
-
Why subject matter expertise is critical for AI development and integration
-
Practical guidance for researchers: choosing reliable AI tools, when to trust outputs, and when to verify
-
For publishers and institutions: the moral obligations of building AI’s “truth infrastructure,” plus where to prioritise investment
Ian Mulvany is BMJ Group’s Chief Technology Officer, driving digital strategy and innovation across products and platforms.
More from Ian
Notes on vibe coding – “what works for me”
Recently, Ian has been deep in vibe coding, building not just small experiments but a substantial app with significant functionality, one he considers as production ready as many tools already in use. His approach balances two modes: striving for robustness versus simply getting things shipped.
On the robust side, Ian has developed a disciplined workflow: over 380 passing pytest tests for endpoints and views, clear design documents in markdown, and clever use of LLMs as coding partners. He leverages Claude for everything from drafting user-facing API docs to pruning codebases for duplication and anti-patterns, and even bounces knotty problems between models to unlock solutions. Git is a constant, Warp’s chat mode powers his commit messages, and Docker containers fuel his throwaway experiments.
Recently, Ian has been reflecting on what he calls FONO (fear of not operating). With LLMs capable of so much and ideas spilling over, he sometimes feels time slipping by without an agent to run with them. As agentic systems mature, that will change, but for now, it’s a fascinating moment to be coding at the edge.
And yes, Ian insists, GPT-5 is a beast.
Related posts
-
Practical lessons for publishers -
Exposing hidden influence in infant feeding advice -
Facing the future, fueling the NHS -
Advancing global healthcare resilience
