Two weeks ago, I gave a talk for the charity Smell Taste about the miraculous chewing gum experience which returned my sense of smell. Just another talk in the life of Paul. Except a throw away joke about Willy Wonka ended up getting quoted in the Telegraph and the Sun, with more outlets to follow (I think LADBible was interested?!)

Now when you google image search me you might find a picture of Violet Beauregarde as an exploding blueberry… but don't they say, "there's no such thing as bad publicity"?

— Paul

We've loved having amazing guests like Dr. Acacia Parks, Dr. Ali Darcy, Dr. Hugh Harvey and more on the show to share their passion and wisdom in the Health and MedTech startup space.

Until the exciting revamped release of Season 2, if you'd like to have our very own Dr Paul Wicks as a guest on your podcast, drop us a line!

He has just recorded a few episodes with Dr. Annabelle Painter for the Royal Society of Medicine's Digital Health Podcast and with Adam Turinas on The HealthTech Marketing Show.

Re-Listen to Season 1 of Prove It! on Your Favourite Platform:

On May 12th, a team at Columbia University led by Maxim Topaz published a sweeping audit in Nature Medicine. They analyzed 2.5 million papers published in the last three years and ran an automated check on every reference.

In 2023: 1 in 3,000 papers contained fabricated citations.
In 2025: 1 in 458.
In the first seven weeks of 2026: 1 in 277.

That's a 10-fold increase in three years. And it's accelerating.

When a paper says "one in three cases of cancer is caused by [X]" and includes a citation, you assume someone did that research. You don't check. No one checks every reference in a paper—the whole edifice of science depends on that assumption. If you can't trust the citations, the entire tower collapses.

Here's the economics: in some countries, universities pay cash bonuses for papers in top-tier journals. China, India, Russia… each Nature publication or NEJM paper brings literal cash money. National pride is involved. Careers depend on it.

At the same time, researchers are drowning. The average clinical trial now generates 6 million data points (up from 2 million a decade ago). Writing up findings has become administratively brutal. And AI promised a shortcut.

So researchers (and increasingly, peer reviewers) started using LLMs like ChatGPT to draft papers, to write peer reviews, and to generate reference lists.

The problem is that language models don't know the difference between real and fabricated. They're pattern-matching machines. If you ask them to cite a paper on PTSD treatment, they'll invent one that sounds plausible rather than admit they don't know.

And here's where it gets genuinely sinister: they've started leaving fingerprints.

In the last 18 months, the word "goblin" has appeared in LLM-generated content with bizarre frequency. So has "gremlin."

Not in isolation. These words appear in generated text at rates that should be statistically impossible.

OpenAI investigated this and found the culprit: their training data includes Reddit discussions about fantasy worldbuilding. When asked to generate scholarly language, the model sometimes defaults to nerd-culture framing. Big Bang Theory energy. When someone gives a thumbs-up to an answer that's clever or funny or culturally aware, the model learns to replicate that pattern.

The goblins and gremlins are canaries.

But here's the terrifying part: what if someone generates a paper claiming to have discovered a new neurotransmitter involved in schizophrenia? What if they hallucinate data from brain scans—data that doesn't exist but sounds medical enough to pass initial scrutiny?

How would we know it's not a goblin?

How would we know we're not redesigning an entire drug development pipeline around a false finding that will block better ideas from ever getting funded?

At the Hardian Summit, I shared this guidance:

In a world where speed is no longer a competitive advantage, the real moat isn't being the first to publish. It's being the only one that people trust.

The Hardian Health Summit was a success—75 people in the room, a mix of founders, regulators, clinicians, and health tech advisors. 

Founders kept asking the same question: "I've got a regulatory strategy and a product roadmap. Where does evidence fit?"

The honest answer: it should have been there from day one. Not as an afterthought. As part of the infrastructure layer that makes everything else possible.

One of the pieces that came out of the summit is a flowchart I've been developing on journal selection and publishing strategy. There are about 15 decision points before you even submit a manuscript. But that complexity is where the real work happens. And if you can systematize it, you move faster and smarter.

This weekend I am taking an unexpected break from the realm of MedTech and HealthTech startups while attending Bearded Theory in Catton Park. I must admit I've been quite excited for this local independent music festival.

We are also less than 1 month away from my favourite HealthTech event of the year… HLTH EU! Will I be seeing you there?