Mediterranean Diet
The evidence behind the most-studied diet in the world — and why the original version of this article was nonsense written by a language model.
Contents
<p>A confession about this page, because the honesty matters more than the tidiness. The article that used to sit here was generated by GPT-2, an early language model, as a deliberate experiment in machine-written text back in 2021. It read plausibly and it was almost entirely wrong — a smooth cascade of invented citations, garbled biochemistry, and confident nonsense about “meat-lover’s fries” and cancers of the “colon, colon, liver.” I’ve kept the URL and rewritten the substance from real sources, partly because the diet deserves an accurate write-up and partly because that old text is a perfect, preserved specimen of why you cannot trust fluent prose to be true prose. If you want the longer version of that lesson, I wrote about the practical limits of running your own models in <a href="/story/local-llms-a-practical-comparison-of-llama-mistral-and-gemma-for-real-work/">a comparison of Llama, Mistral and Gemma for real work</a>; the short version is that a model’s confidence and a model’s correctness are entirely unrelated quantities. So here is the actual Mediterranean diet, with names, dates and studies you can check.</p>
<h2 id="where-the-idea-came-from">Where the idea came from</h2><div class="ad-unit ad-in-article" aria-label="Advertisement">
<span class="ad-label">Advertisement</span>
<ins class="adsbygoogle" style="display:block;text-align:center"
data-ad-client="ca-pub-3726833845844946"
data-ad-slot="3291553914"
data-ad-format="auto"
data-full-width-responsive="true"></ins>
<script>(adsbygoogle = window.adsbygoogle || []).push({});</script>
</div>
<p>The “Mediterranean diet” is not an ancient folk tradition marketed at you — it’s a scientific observation with a specific origin. In the years after the Second World War, the American physiologist <strong>Ancel Keys</strong> noticed that heart disease, then surging in the United States, was strikingly rare among the rural populations of southern Italy and the Greek island of Crete, despite their being far from wealthy. That observation grew into the <strong>Seven Countries Study</strong>, launched in 1958, which followed roughly 12,000 middle-aged men across cohorts in the United States, Finland, the Netherlands, Italy, the former Yugoslavia, Greece and Japan, tracking their diets and their hearts over decades.</p>
<p>The finding that made the diet famous was the contrast between the extremes. The Finnish and American cohorts, eating diets heavy in saturated animal fat, had by far the highest rates of coronary heart disease. The Cretan cohort, eating a diet built on olive oil, bread, vegetables, fruit, legumes and fish, with very little red meat, had among the lowest — and lived long doing it. Keys didn’t invent the diet; he documented one that already existed on Crete in the 1960s and connected it to cardiovascular health.</p>
<h2 id="what-the-pattern-actually-is">What the pattern actually is</h2>
<p>It helps to describe the diet by what’s on the plate rather than as a brand, because the branded, ultra-processed “Mediterranean” products in a supermarket bear little resemblance to what was studied. The traditional Cretan pattern that Keys recorded was, roughly:</p>
<ul>
<li><strong>Olive oil as the principal fat</strong>, not butter or lard. In the studied populations, a large share of daily calories came from unsaturated fat, mostly olive oil.</li>
<li><strong>Plants as the base of every meal</strong>: vegetables, fruit, whole grains, legumes, nuts.</li>
<li><strong>Fish and seafood</strong> regularly; <strong>poultry, eggs and dairy</strong> (often as cheese and yoghurt) in moderation.</li>
<li><strong>Red meat rarely</strong> — a few times a month, not a few times a day.</li>
<li><strong>Wine in moderation</strong>, typically with meals, in the cultures where it featured.</li>
</ul>
<p>Note what’s absent: this is not a low-fat diet. It’s a diet that swaps saturated fat for unsaturated, and refined carbohydrate for whole plants. That distinction is the whole point, and it’s exactly the part the old AI-generated version of this page mangled into gibberish about “grinding meat down with water.”</p>
<h2 id="the-strongest-modern-evidence--and-its-asterisk">The strongest modern evidence — and its asterisk</h2><div class="ad-unit ad-in-article" aria-label="Advertisement">
<span class="ad-label">Advertisement</span>
<ins class="adsbygoogle" style="display:block;text-align:center"
data-ad-client="ca-pub-3726833845844946"
data-ad-slot="3291553914"
data-ad-format="auto"
data-full-width-responsive="true"></ins>
<script>(adsbygoogle = window.adsbygoogle || []).push({});</script>
</div>
<p>The observational Seven Countries Study could show correlation but not prove cause. The landmark attempt to test the diet as a genuine experiment was <strong>PREDIMED</strong>, a Spanish randomised controlled trial whose main results were published in the <em>New England Journal of Medicine</em> in 2013. It enrolled 7,447 people at high cardiovascular risk and randomly assigned them to one of three diets: a Mediterranean diet supplemented with extra-virgin olive oil, a Mediterranean diet supplemented with mixed nuts, or a control low-fat diet. The two Mediterranean groups saw their rate of major cardiovascular events — heart attack, stroke, or cardiovascular death — fall by roughly 30% and 28% respectively compared with the control group.</p>
<p>Here is the asterisk, and it’s important precisely because I just spent three paragraphs telling you not to trust confident prose. In 2018 the original 2013 paper was <strong>retracted</strong> by the <em>New England Journal of Medicine</em> after it emerged that some participants hadn’t been properly randomised — a portion had been enrolled by household or by clinic rather than individually. The authors reanalysed the data accounting for the flaw and <strong>republished</strong> it the same day, and the corrected results still showed the same benefit. So the finding survived, but the episode is a useful reminder that even peer-reviewed, headline nutrition science gets things wrong and has to correct itself. That’s science working, not science failing — but it means “a 2013 study proved it” is a sloppier claim than “a 2013 trial, corrected in 2018, supports it.”</p>
<h2 id="why-it-plausibly-works--the-mechanisms-kept-honest">Why it plausibly works — the mechanisms, kept honest</h2>
<p>It’s worth being clear about what we understand of the <em>why</em>, and equally clear about where the understanding runs out. Several mechanisms are well supported. Replacing saturated fat (butter, lard, fatty red meat) with monounsaturated fat from olive oil lowers LDL cholesterol, and extra-virgin olive oil in particular carries polyphenols with measurable anti-inflammatory and antioxidant activity. The high intake of vegetables, legumes, nuts and whole grains delivers fibre, which improves blood-lipid profiles and blood-sugar control. Oily fish supplies omega-3 fatty acids associated with better cardiovascular outcomes. A PREDIMED sub-analysis found that each additional 10 grams a day of extra-virgin olive oil was associated with roughly a 10% lower cardiovascular risk and a 7% lower mortality risk — a specific, checkable figure, not a vibe.</p>
<p>But the intellectually honest position is that nobody has cleanly isolated <em>which</em> component does <em>how much</em> of the work, because you can’t easily un-tangle a whole dietary pattern into single variables. There’s even a long-standing argument — raised in analyses of the Cretan cohort specifically — that part of what Keys measured on 1960s Crete wasn’t just the food but the periodic religious fasting of the Orthodox calendar, effectively intermittent caloric restriction layered on top of the diet. That doesn’t undermine the pattern; it complicates the tidy story. Real nutrition science is messier than a magazine headline, and anyone who tells you they’ve pinpointed the single active ingredient is selling something.</p>
<h2 id="the-criticisms-worth-taking-seriously">The criticisms worth taking seriously</h2>
<p>Two fair critiques keep the enthusiasm grounded. First, Ancel Keys has been accused, with some justification, of building his hypothesis on a selective reading — the “Seven Countries” were the seven he chose, and later commentators pointed out that a wider sample would have muddied the neat fat-versus-heart-disease line. His conclusions broadly held up under decades of subsequent work, but the original study design had its thumb closer to the scale than the legend admits. Second, and more practically, the diet studied on Crete in 1960 was inseparable from a way of life — physical labour, small portions, home cooking, near-zero ultra-processed food. Lifting the food list out of that context and dropping it into a sedentary modern life with a car and a supermarket may not reproduce the same benefit. The pattern is not a pill.</p>
<h2 id="what-this-does-and-doesnt-mean-for-you">What this does and doesn’t mean for you</h2>
<p>The reasonable, evidence-grounded reading is this. The Mediterranean <em>pattern</em> — lots of plants, olive oil instead of butter, fish over red meat, whole grains over refined ones, not much ultra-processed food — is one of the best-supported dietary patterns we have for cardiovascular health, backed by both long observational cohorts and randomised trials. It is not a magic wand, it will not “burn fat from your cells,” and no single food in it is a cure for anything. The benefit is in the whole pattern followed over years, not in buying a bottle labelled with an Italian flag.</p>
<p>What it emphatically is <em>not</em> is the free-floating collection of half-remembered claims the previous version of this page served up. If you want to actually cook this way rather than read about it, the practical barrier isn’t knowledge, it’s organisation — which is why I keep my own recipes in <a href="/story/mealie-a-self-hosted-recipe-manager-for-people-who-cook/">a self-hosted recipe manager</a>, tagged so I can pull up the vegetable-and-legume-heavy, olive-oil-based meals on a weeknight without hunting.</p>
<p>In concrete terms, “eating this way” is less dramatic than a branded diet plan and more about steady defaults. It means olive oil as your everyday cooking fat instead of butter. It means a lentil, chickpea or bean dish two or three nights a week where you’d otherwise reach for mince. It means fish once or twice a week, a bowl of fruit where a snack would go, whole grains standing in for the refined ones, and red meat becoming a genuine occasion rather than a default. Nuts and yoghurt do a lot of quiet work. None of it requires imported speciality products or a subscription; the studied version was, after all, a peasant diet built from whatever grew locally and cheaply. If anything, the most “Mediterranean” move you can make is to cook more of your own food from plants and buy less of it pre-made, which is exactly the part a marketed jar of “Mediterranean-style” sauce inverts.</p>
<p>And crucially, it is a <em>pattern to drift towards</em>, not a set of rules to pass or fail. The evidence is about long-run habits, so the useful question isn’t “am I doing the Mediterranean diet correctly this week” but “over a year, is my plate moving in this direction”. That framing is both kinder and, going by the studies, more honest about how the benefit actually accrues.</p>
<h2 id="the-real-point-of-keeping-this-page">The real point of keeping this page</h2>
<p>I could have quietly deleted the old machine-generated article and pretended it never existed. I’ve left this note instead because it’s a small, concrete case study in a problem that has only grown since 2021. A language model produced text about the Mediterranean diet that was grammatically flawless, structurally convincing, studded with fake-looking citations like “(Carmen et al., 2005)” — and factually worthless. Five years on, models are vastly more capable and the failure mode is subtler, not gone: fluent, authoritative, and wrong is still a thing they do every single day, and it is now considerably harder to spot precisely because everything else about the output has improved so much. The lesson from a 2021 GPT-2 experiment is the same lesson that matters when you’re running a modern model on your own hardware: verify the claims, trace the sources, and never mistake confidence for correctness. The diet is real, the evidence is real, and the only way to know that is to check it against named studies and dates — which is precisely the one thing the machine that wrote the first draft of this page never did, and never could.</p>
Advertisement
Related Content
Advertisement
