Èter: I and AI review one of Madrid’s 32 Michelin-starred Restaurants

by Andrew Chalk

On a vacation to Madrid this summer I had the opportunity to eat dinner at Èter, a restaurant with one Michelin star situated in the leafy Delicias neighborhood about a mile and a half directly south of The Prado. 

In a world where Michelin-starred restaurants are architectural statements in a country that has been hegemonic in art for a century, the exterior of  Èter has all the impact of a real estate agent’s sub-office in a drowsy suburb. We really did almost miss it on our walk from the Uber to the curb!

Èter is a bijou-sized room with only about a dozen tables. The chef and the sommelier are the founding brothers, Sergio and Mario Tofé, six years apart in age, and with a reputation in Madrid before they broke out on their own to found Èter in 2020. They serve a seasonal tasting menu that was E125 when we visited (wine pairing flights are also available).  

The 'Carpus' menu served over summer. The original Spanish is on the left and Google Translate into English on the right.

Technical Diversion- AI Comes To Restaurant Reviews

Like many people I use a voice recorder to capture the details of the dish descriptions during service. It turns out this was especially important at Èter as this restaurant provided more (welcome) chef involvement for I and The Moll than any restaurant since I went to Georgie in Dallas shortly after RJ Yoakum’s promotion to executive chef. 

Sergio himself conveyed to us the overarching mission of Èter, and in the cozy room with constant background murmur, and the challenge of accents in translation, I was rescued from catching so little as to look caught like a deer in the headlamps by a remarkable tool that is an early product of the AI age. It is the Plaud Pro, a voice recorder with AI. I paid just under $200 retail for it. It is so important that it deserves explanation. I should say up front that I have no commercial interest in the company that produces it. 

To understand the Plaud Pro consider the normal use of an ordinary voice recorder.

Stage 1: In the restaurant, hold the recorder near the speaker. Turn on and off to record dish descriptions. The result is recorded files of each. Background noise has a critical effect on the quality of the recordings.

Stage 2: After the meal, download the recordings and either manually transcribe them or feed them through a transcription service. The result is a text representation of each audio file. 

Stage 3: Weave the transcripts into the review. Hopefully, the quality of the audio and the time with the chef are long enough to generate a really informative result for the reader. Ideally, clearly expressing the chef’s mission underpinning the concept, menu, dishes, beverages, and  service choices that represent the story behind the meal.

With Plaud, it’s a transformational shift. Plaud’s AI creates recordings of such astounding quality you can almost use them unedited. But it doesn’t just give you a recording, it creates the transcript in seconds (minutes for a very long one). Then, it takes the transcript and creates what it calls a “Summary”. But that word understates what the summarisation stage is. The best way to explain it is with our first course at Èter as an example.   

Here is the course, it has just been put down in front of us, we don’t know what it is. The charming waitress knows her stuff and gives a detailed explanation that is virtually inaudible in the recording below due to background noise and an accent. 

AI Recorder Output

Here is an image of Plaud’s output. There are two tabs. First, the transcript tab.

 At the top is the recording, below it is Plaud’s transcript. I did not edit the transcript you see here. The speech recognition quality is uncanny, albeit not yet perfect.

Next is the Summary tab. There is so much information it took up three screen grabs. Be aware that it takes a couple of reads to realize how far AI has gone with this. To repeat, it still makes errors! They are sometimes misleading, sometimes confusing, sometimes hilarious. 

Those are screen dumps, the actual Plaud output is here. When you play the recording, you see how well Plaud recognized it. 

While Plaud can produce a much more accurate account of what the restaurant staff said, it is still up to the reviewer to distill his own comments on the food. In this course, these were three excellent appetizers with the Indian flavors of the panipuri topping out in my estimation. 

Highlights

Hiramasa – Watermelon – Umeboshi

The Kingfish with Watermelon Dashi and Pickle Vinaigrette was a hearty, umami-rich mouthful of sashimi accented by borage flowers and mounted on a base of Koshihikari rice. I loved the subtle contrasts of smoke, pickle and dashi although this was a very Japanese cuisine-inspired dish.

AI Output

Salmorejo - Smoke - Ham

It looks like Gazpacho and Salmorejo, from Cordoba, is pitched as a forerunner of that soup. A modern twist is transforming the traditional garnish of cured ham into cured ham ice cream. This innovation worked and despite this being the only dish to return on every summer menu I would not get tired of it.

AI Output

Swiss chard – Tendon - Huitlacoche

The next course sat rather ominously as a dark hunk on the plate. The earthy tastes were the unleashing merits.

AI Output

Beverages

I have excluded discussion of wine and other beverages served with the pairing. This doesn’t do Mario’s contribution justice. His pairing flight dotted into nooks and crannies all over Europe, with many selections (e.g. Crémant du Jura, Txakolina Basque Wine, Vin de France Pinot Noir from Henri Chauvet, belying its humble appellation heritage, and tart Madeira) off the mainstream path. Interspersed with formidable mainstream selections (e.g. Champagne Ruppert-Leroy Brut Nature).

Later Courses

As the meal progressed the first-principles thinking of the first four courses played out. A

mexican tamal, with a moutard sauce and epazote pesto that was alluringly rich in the mouth. The Champagne pairing idea was a winner as well. And it was the first time I had tasted an epazote pesto. It was fragrant and distinctly unbasil-like.

The Mexican tamal, with a moutard sauce and epazote pesto.

The chef here is unbound by convention. He picks a guideline: locality, and then combines without regard to how ingredients were combined before. It is reminiscent of Tom Sellers’ Restaurant Story.

Plumbing this with an AI recorder was an experiment and I am left unsure of the best way to use AI in the future. Plaud is staying due to its ability to record and transcribe in one operation, but those summaries detract from a linear content flow. Maybe they should go into their own compartments. TBD.

But Èter? Recommended without hesitation to serious food explorers.