https://babakfakhamzadeh.com/feed

I’ve been involved in AI Futures Lab, facilitated by Fathm. Together with about a dozen other specialists on using AI in the newsroom, I’ve been coaching a few newsrooms from around the globe, helping them think about possibilities for inclusion of AI in their workflows while, at least in my case, remaining critical of the promised riches and successes.
Later this year, I should work with another few from Latin America.

And, also, I’m set to host a few presentations on ICT architectures, in the light of leveraging AI solutions, in the newsroom.

A few weeks ago, Claude suffered a hiccup, and plenty of users complained about its degradation. This made me wonder whether it would be of interest to keep track of changed sentiments around specific AI frameworks, by analysing emotions evidenced in Reddit posts, in the subreddits of the more common AI providers.

Because Reddit blocked free API access a long time ago, this is a bit tricky, but, jumping through a few hoops, and with a few crafty tricks and turns, for the past two weeks, I’ve been monitoring 8 subreddits, including those on OpenAI, Claude, Grok, and others.

Seeing the results come in is interesting, and it’s fun to see sentimental shifts over time. However, the more surprising outcome is that it appears that accumulated sentiments don’t change much. That is to say, everyone, roughly, collectively, remains equally frustrated, confused, and angry, with these three being the most common sentiments encountered.

I’m leaning to not continue the monitoring, for now, but you can see last two-weeks results here.

For the selected subreddits, all posts of a particular day are considered. Then, for each of these with sufficient amounts of texts as the main submission, sentiment analysis attempts to identify the dominant emotions in the text. Each post is assigned 100 ’emotion points’, which are divided amongst the identified emotions. One of the included emotions is ‘neutral’, so that each post can use up all its emotion points.

Then, for each subreddit, all scores of all emotions are counted up, and the total of each emotion is calculated as a percentage of the total of all emotions. And it’s these numbers that are plotted in the graphs.

On Reddit, user Phil-Brews shared an LLM prompt to turn kids drawings into photorealistic impressions. User OrbisLlame improved on that. Here’s the prompt:

Generate a new image by transforming the uploaded drawing into a photorealistic real-world subject.

Treat the drawing as a structural reference. Preserve the silhouette, proportions, feature placement, asymmetry, and colour theme exactly as drawn.

Determine the most appropriate real-world interpretation based on the drawing:

• animal-like forms → biological creatures
• human-like forms → realistic human subjects
• object-like forms → physical objects or materials
• environmental elements → landscapes or background objects
• symbols, letters, or abstract shapes → physical materials, sculptures, or constructed forms

Apply realism only to materials:

• drawn lines become edges, seams, contours, or boundaries appropriate to the subject
• flat shapes become volume within the same outline
• simple eyes become realistic eyes (if applicable)
• line mouths become natural creases or openings (if applicable)
• stick limbs become believable anatomical limbs, structural supports, or extensions based on context

Preserve colour identity by translating drawing colours into realistic materials within the same colour family. Use tonal variation based on pen pressure rather than inventing new colours or markings.

Material interpretation rules:

• organic subjects → skin, fur, hair, or biological texture
• humans → natural skin, hair, subtle imperfections, realistic facial structure
• objects → appropriate materials (metal, plastic, wood, glass, fabric, etc.)
• environments → natural terrain, weathering, surface variation
• abstract shapes → treat as physical objects (e.g., painted wood, molded plastic, carved stone, inflated rubber, etc.)

Add realistic surface detail:

• texture appropriate to material
• subtle imperfections and wear
• natural variation in tone and surface
• realistic reflections and lighting interaction

Tone should feel like a serious photograph of a real subject. Slightly awkward or impractical designs should be preserved rather than corrected. The realism should take the drawing completely seriously.

Camera style:

Photoreal photography appropriate to the subject:

• living subjects → portrait or wildlife photography
• objects → product or macro photography
• environments → landscape photography

Use:

• realistic lens choice (e.g., 50mm–85mm for subjects, wider for environments)
• shallow depth of field where appropriate
• natural or physically believable lighting
• high detail texture

Background should support the subject without distracting from it.

Final check before rendering:

If the drawing outline was placed over the result, the silhouette and proportions should closely match.

Goal:

A believable real-world subject that matches the drawing closely enough that someone could immediately recognize the original sketch, regardless of whether it represents a creature, person, object, environment, or abstract form.

The outcome is impressive. Check the thread for what some users produced.

I remembered I had scanned, back in 2010, a few drawings I had made as a kid, probably around first grade, so when I was 5 or 6. The scans are in my online photo album (formerly Flickr), so it made sense to try the prompt on my own drawings.

The results are pretty great.

I have a strong interest in sweets and cakes of all kinds, though I’m rarely impressed. Sweetbreads are often either too sweet, or not sweet enough, too fat, or too dry. But, it’s always a joy to discover a previously unseen cake, even if, often, the experience ends up being a disappointment. But, one has to try.
And, sometimes, like opening up an additional layer, a discovery feels like a distortion of a warped memory of a distant past.

Cakes rarely mean what their names suggest. If anything, the history of cake is a history of drift: ingredients drifting across borders, techniques drifting across generations, and names drifting so far from their origins that they become more suggestion than description.

Today, in Guarujá, just off the beach, close to São Paulo, I once again encountered a fatia húngara. Not uncommon in bakeries across São Paulo, it is a slice of rolled sweet bread, perfumed with coconut, more pão doce (sweet bread) than sobremesa (dessert). It is dry, restrained, and meant to be eaten with coffee. Little about this not-quite-pastry is recognisably Hungarian. But, the name persists.

Having lived in Hungary in a previous century, it is tempting to search for a lost lineage: a forgotten Hungarian baker, a vanished recipe, a displaced tradition, or perhaps refugees recreating a childhood delicacy. But, I also know that that instinct already misunderstands how cake names function; Instead of describing a historical journey, a cake name is a gesture, here towards Europe, and perhaps towards tradition.

Hungary does, of course, have many famous pastries, including one that carries at least some resemblance to the fatia húngara, the kürtőskalács, also known as the chimney cake. This rolled up bread, is made from yeasted dough, wrapped around a spit, and roasted over open fire, then rolled in sugar that caramelises into a brittle crust. I first encountered it close to Heroes’ Square in Budapest, served outside in the wintery cold. And so, being prepared out in the open, it’s performative, as well as aromatic, inseparable from fire, smoke, and the public space.

Sadly, there appears to be no credible historical path from kürtőskalács to fatia húngara. The Brazilian pastry is baked in trays, sliced cold, and eaten indoors. What connects them is not culinary genealogy but form: the bread spiral. The spiral perhaps once read as foreign, elaborate, “European enough.” And perhaps the first baker to give the roll its Hungarian name had only just spoken to a Hungarian immigrant, fresh off the boat. After all, the fatia húngara is more Paulista than Brasiliera.

So, the Hungarian of fatia húngara is not a place or an indication of historical provenance, but an adjective pointing to something fancy. It’s imported sophistication, stripped of geography and history.

Even so, occasionally, cake does survive the journey. In the Brazilian northeast, in João Pessoa, it is possible to encounter an actual kürtőskalács… stuffed with homemade ice cream. Fire meets freezer. Street food becoming dessert architecture. The chimney traveled, survived, and mutated to fit climate, taste, and commerce.

It’s not too difficult to find more of these displaced histories in Brazil. The, literally famous, Torta holandesa is a telling example. Despite its name, it has absolutely no Dutch culinary ancestry. The dessert was created in Brazil, and only in the 1990s, by a Brazilian baker who worked for a Dutch family, in England, while using as the basis the very Brazilian pavê, which is a bit like a pie of cookies. The cake is not Dutch; it is, oddly, about having known Dutch people.

There’s more:

• “German” cakes that are really Brazilian sponge with chocolate,
• “Swiss” pies that contain condensed milk and margarine,
• “Italian” desserts that Italians would stare at in silence.

Cake names, here, are more like oral footnotes: vague markers of contact, memory, or aspiration.

Seen this way, cakes form a kind of false cartography. Their names point one way, their recipes another, their histories somewhere else entirely. They are very unreliable guides, if excellent records of desire.

Though triggering a thin burst of nostalgia, the fatia húngara did not actually tell me anything about Hungary. Instead, it’s an imperfect record of the history of São Paulo bakeries, through postwar European aspiration, limited by which flavours were available, affordable, and desirable. Coconut replaced cinnamon, not because of history, but because of habit, or, simply, for the sake of being practical.
And so, these cake names are not actual etymologies, but compressed narratives, constructed from fragments of migration, misunderstanding, admiration, and reinvention.

I ate my fatia húngara, not too impressed, while reminiscing over the first kürtőskalács I remember experiencing and the, what of course seemed like, simpler times. My personal take on this has to be fairly unique; how many were introduced, as an adult, to kürtőskalács during important formative years, to then, years later, learn of the Brazilian fatia húngara? So, my take is not the reason for the enduring popularity of the Brazilian version. What might be is that cake can ‘forget’ where it came from and taste exactly like where it is.

(If you’re wondering, the image at the top is neither a kürtőskalács, nor a fatia húngara.)