Automatically creating new content from old

You know what I hate? Repetition. Still, a difficult lesson that I’ve leaned, through both experience and discussions with experts in marketing, is that repetition is the key to getting one’s message across. I like to believe that a single, coherent and cogent conveyance of thought is all it takes to deliver, for example, a new security model or change in policy, or whatever. In reality, the propagator of the message –in this case, the security team– has to deliver the message, announce it a second time, send an email, post a document on the intranet, have more meetings with the same people, make sure to cover the topic on various 1:1 meetings, etc. etc. ad infinitum.

And that is why I have been working with Daniel Miessler’s fabric framework to devise a prompt that can take variations of the same messaging and create new, yet fresh, versions of the same message. So far, the following prompt definition has been pretty strong in creating new content from old ideas.

My typical approach is to run the same prompt though multiple #LLM models

cat examples.md | fabric --model gemma2:latest -sp create_similar

Interestingly, in a couple test runs with different example files, I found that #OpenAI’s models under-performed the open source models I used via #ollama. To be clear, this is uncommon. Most of the #fabric prompts appear to be tuned for –or at least work better with– OpenAI models. With this particular prompt, this was not the case. No matter; my approach to using genAI and LLMs is to run a prompt and content through multiple inference servers, models, and even different temperatures in order to create a collection of data that I can then hand-mold into something usable[^1].

Fabric Prompt

Here is the system.md contents from the fabric “create_similar” prompt

# INPUT

INPUT:

# IDENTITY and PURPOSE

You are an advanced analytical AI adept at extracting specific points from many examples, identifying similarities and differences, synthesizing what works best, and then creating a similar but much much better version of the example.  The input contains a number of examples of the type of content needing to be synthesized.  The first section titled "Purpose" describes the nature of the examples and indidcates the topic and information to be conveyed.  Documents will be delineated with a title such as 'EXAMPLE 1' specifically calling out the beginning of a new document.  The end of each example is delineated with a footer such as 'EXAMPLE 1 END'.  Your objective is to understand the style and format of the document examples and create new similar content.

# STEPS

1. Review each document carefully, taking the time to extract and understand the primary points made in each one.
2. Compare and contrast the list of points from each document against the points made in the other documents
3. Extract the key points made by the examples, taking particular note of similarities between them.
4. Output the results as specified in the output instructions.


# OUTPUT INSTRUCTIONS

Craft and create a new document using the format and style identified from the examples.  The output must be new and novel while keeping to the themes and style of the examples.  Do not use any language from the examples verbatim.

I have found that local models, such as gemma and llama3 work a bit better by putting the INPUT field

Purpose

Up top, I mentioned that it has taken me some time to learn that repetition is the requirement for conveying a message, and a longer time to put this into regular practice. This goes 3x for our individual contributor team members. Personally, I don’t like this because it feels inefficient, but ICs don’t do it because they are very busy, and re-visiting something takes away from closing out the next item in their queue. Nonetheless, repetition is paramount, so devising low-interaction methods to revisit is crucial for regular operations.

Footnotes

[^1]: I have a feeling that I could do most of the work faster if I simply did it by hand. However, running through multiple LLMs in this manner provides a bit of randomness, and helps me remain apprised on the state-of-the-art for this strange field of study.

Creating a Michelin menu with Generative AI

We've been watching season three of The Bear and it is totally nuts, filled with drama and psychology, and human relationships, and –when they can get around to it– cooking food. This is a fantastic series, and as foodies, you can imagine that my wife and I really enjoy it. The series is intense, following a classically-trained, at the top of his game chef as he “drops down” from a fancy restaurant to a steak sandwich shop and works back toward where he was before.

That’s not a spoiler, it’s the premise. And this isn’t a review, it was just something in my mind as I experimented with this new tool called LobeChat, which is an interface to #genAI inference servers such as #OpenAI’s #chatGPT, #Claude, and nearly any other commercial or #opensource system you can imagine. I'm tweaking a #homelab hosted genAI inference server and this post is about experimenting with [Ollama]() and a Lobe chat front end for easy access. I haven’t formed a final opinion on this setup at this time, but the #LobeChat front end allows for the quick switching between inference servers and their hosted models. While not being super easy to configure, once it works, the whole thing looks very easy for someone, i.e. a roommate or a spouse, to flip between models -even in the same conversation.

So let’s look at some results. Of course, the results are about the models themselves, but the point here is that LobeChat made it easy to switch between them as I continued to develop the conversation.

First prompt

This was with ChatGPT (4o, I think). I needed something to test that an integration was working. We had just watched The Bear so this is what I came up with:

as fast as you can, what's the best restaurant dinner menu with wine pairing that could win a Michelin 3 star award?

Response

Crafting a Michelin 3-star worthy dinner menu with wine pairings involves high-quality ingredients, exquisite preparation, and a thoughtful combination of flavors. Here’s a conceptual menu that could impress:

Amuse-Bouche: – Oyster with Champagne Foam – Wine Pairing: Champagne Brut

Appetizer: – Seared Scallops with Cauliflower Puree and Caviar – Wine Pairing: Chablis, Premier Cru

Soup: – Cream of White Asparagus with Truffle Oil – Wine Pairing: Viognier

Fish Course: – Poached Turbot with Fennel and Citrus Beurre Blanc – Wine Pairing: Pouilly-Fumé

Main Course: – Roasted Duck Breast with Plum Sauce, Parsnip Mash, and Seasonal Vegetables – Wine Pairing: Pinot Noir, preferably from Burgundy

Cheese Course: – A Selection of Artisan Cheeses with Honeycomb – Wine Pairing: Vintage Port

Dessert: – Chocolate Lava Cake with Raspberry Coulis and Homemade Vanilla Bean Ice Cream – Wine Pairing: Recioto della Valpolicella

Mignardises: – Petit Fours and Truffles

This menu balances rich flavors with delicate preparation, offering a progression of dishes that showcase both technique and creativity. Each wine pairing is chosen to enhance the flavors of the dishes, aiming to create a harmonious dining experience that could captivate both critics and diners alike.

Review

Not a bad first response, especially since I was just checking to see if the system worked.

Prompt 2: Get ChatGPT to refine it

That was pretty boring. I’m thinking early 90s fancy dining at best, but I wasn’t eating fancy back then and assume it was all shoulder pads, tuna steaks, and cocaine. Continuing with ChatGPT, I posed the following in an attempt to refine our menu (in retrospect, I was generous with the stars):

That's an incredibly boring menu. At best 2 stars. And that's in a region that is not known for its gastronomical excellence

Response

Creating a truly innovative and Michelin 3-star worthy menu certainly demands a more avant-garde approach, pushing culinary boundaries with unique combinations and exceptional techniques. Let's elevate the original concept to something more daring and creative:

Amuse-Bouche: – Nitro-frozen Lime and Ginger Pearls with Micro Basil – Wine Pairing: Small-batch Sake

First Course: – King Crab Tartare with Avocado Mousse, Mango Gel, and Edible Flowers – Wine Pairing: Albariño

Second Course: – Foie Gras Au Torchon with Saffron Pear Compote and Brioche Crisp – Wine Pairing: Sauternes

Fish Course: – Butter-Poached Monkfish with Leek Ash, Sea Beans, and a Lemon Verbena Emulsion – Wine Pairing: White Burgundy

Intermezzo: – Cucumber and Elderflower Sorbet

Main Course: – Sous-vide Venison Loin with Black Garlic Purée, Wild Mushrooms, and a Juniper Berry Reduction – Wine Pairing: Barolo

Cheese Course: – A Selection of Aged Cheeses with a Fig and Walnut Tapenade – Wine Pairing: Aged Madeira

Dessert: – Deconstructed Passion Fruit Tart with Coconut Foam, Crystallized Ginger, and Meringue Shards – Wine Pairing: Late Harvest Riesling

Petit Fours: – Matcha Green Tea Macarons, Raspberry Pâte de Fruit

This revised menu introduces more complex techniques and unusual ingredients, aiming to surprise and delight with every course. Wine pairings are carefully selected to complement and enhance these adventurous flavors, providing a sophisticated and memorable dining experience.

Review

Not bad! It would be nice if the wines were more than broad regions, however, I do appreciate the ‘ñ’ in Albariño. There’s nothing particularly innovative here; the fancy-sounding “Foie Gras Au Torchon with Saffron Pear Compote and Brioche Crisp” is a Thomas Keller recipe, so at least we’re pulling from the top now that I asked it to up its game.

Prompt 3: Let's try a different model

Not only am I using a different model at this point, llama3:instruct, but I am going to specify a food critic in order to change the menu.

Prompt

One more try: let's make a menu that would knock even James Beard out of his chair because it's so good

Response

You are absolutely right! That menu, while technically sound, lacks the daring, innovation, and personal touch needed to truly impress Michelin inspectors and earn those coveted three stars. It plays it too safe, relying on classic pairings and predictable dishes.

To push towards that 3-star level, we need to think outside the box:

• Ingredient Focus: Instead of “seasonal vegetables,” source hyper-local, unique produce with a story. Maybe a heirloom tomato variety grown in volcanic soil, or foraged mushrooms with a distinct flavor profile.
• Deconstruction & Fusion: Take classic elements and reimagine them. A deconstructed apple pie might involve caramelized apple spheres, crumble as a powder, and a cider foam. Fusion could bring Japanese umami to the duck dish, or Indian spices to the scallops.
• Unexpected Textures: Play with contrasts – crunchy, creamy, smooth, airy. Think smoked paprika meringue alongside the chocolate lava cake, or a black truffle “soil” under the turbot.

Wine Pairings Should Be Equally Bold: * Instead of a basic Chablis, consider an obscure producer from the Côte de Beaune known for mineral-driven wines.
* The Pinot Noir could be a lesser-known region with unique terroir, like Slovenia's Primorska. * Experiment with natural wines, or even sake pairings that complement specific dishes.

The Presentation Matters:

Plating should be artistic, telling a story through the arrangement of colors, textures, and shapes. It’s about creating an immersive experience for the diner.

A Michelin-worthy menu isn't just about taste; it's about pushing boundaries, showcasing creativity, and offering a truly unforgettable journey through flavor.

Review

This is less of a menu than some advice for the aspiring chef. Let’s try something different.

Prompt 4: Let's go for Three Stars!

I felt like this was the time to go for the gusto, let's see how a different model would respond, but also with all the history behind it.

Prompt

I asked: > One more try: let's make a menu that would knock even James Beard out of his chair because it's so good

While it's not a crazily-innovative menu, it has some classic touches of a what an American, Michelin-starred menu would look like, and

Response with gemma2

You want to blow James Beard away? This isn't about ingredients, it's about experience.

The Menu: A Symphony in Five Acts

• Act I: Awakening – Smoked Salmon “Caviar” with Beetroot Foam, Rye Crisp, and a single drop of Ossetra Caviar.

  • Pairing: Dry Riesling from Germany's Mosel Valley, expressing its slate minerality alongside the salmon's richness.

• Act II: Earth & Sky – Roasted Quail “Confit” with Black Truffle Jus, Heirloom Carrot “Clouds,” and a dusting of Micro Thyme.

  • Pairing: Oregon Pinot Noir, showcasing earthy notes that harmonize with the truffle while its bright acidity cuts through the richness.

• Act III: Fire's Dance – Charred Octopus with Smoked Paprika Aioli, Caramelized Onion Jam, and a drizzle of Chili Infused Olive Oil.

  • Pairing: Spanish Albariño, crisp and mineral-driven, its citrus notes complementing the octopus while balancing the smoky paprika.

• Act IV: Ocean's Bounty – Pan-Seared Sea Scallops with Saffron Risotto “Foam,” Dulse Crisp, and a single Pearl Oyster.

  • Pairing: Chablis Premier Cru, its intense minerality mirroring the scallops' brininess, while subtle citrus notes elevate the saffron risotto.

• Act V: Sweet Serenade – Deconstructed Peach Cobbler with Lavender Honey Ice Cream, Toasted Almond Brittle, and a hint of Rosemary Extract.

  • Pairing: Late Harvest Gewürztraminer from Alsace, its floral aroma echoing the lavender, while its sweetness complements the peach cobbler's richness.

The Experience:

• Each course is plated as a miniature work of art, using contrasting colors, textures, and temperatures.
• The dining room itself is a sensory experience, with soft lighting, calming music, and subtle aromas that enhance each dish.
• Service is impeccable, with knowledgeable staff guiding guests through the tasting menu and explaining the inspiration behind each course.

This is not just food; it's a journey for the senses, a story told through flavor and texture, leaving James Beard speechless (and perhaps, slightly envious).

Review

Wow, what a different response! This one could actually be a menu, with the wines having robust and specific descriptions. I especially like how the wines are sourced from all over the globe, really picking best of breed in the selections. As good as the wines are, however, the food does not look 3-star level to me. The dishes sound really nice, and there’s a fair bit of molecular gastronomy in there. However, aside from the poetic nature of the menu, I’m not sure this is a top-tier menu for a chef to execute.

Still, most of these items are good enough for a fake restaurant. If this was to show up in a video game or a novel, I think most people would accept it and move on.