#002302 – 17 de Setembro de 2025

Comecei a escrever um diário, offline. Estou a escrever um livro de contos e a procurar recuperar o tempo perdido, aqui. Não é setembro de 2025, devo bastantes textos a esta página web. Mas eles vão saindo, excreções relutantes de organismo ainda adoentado.

OK, maybe not phonetically. But for a museum dedicated to open access, a decentralised, non-algorithmic social media platform seems an obvious place to make ourselves available and claim a space for the national Danish art collection.

A mastodon (well, some sort of mammoth) figurine visiting SMK

Last week, on an early spring Friday, we took our first baby steps into the Fediverse, the collection of platforms connected by the ActivityPub protocol. We did this by setting up an account on the cosy Danish expressional.social server populated by friendly-seeming natives (with an endearing love of image alt texts).

Now, that previous paragraph contains the words ‘protocol’ and ‘server’ and admittedly the Fediverse does require some acclimatisation: it’s just a slightly more abstract concept than your average centralised service. But such is the price of openness and flexibility. When you can “do” the Fediverse almost any way you please, choose your own server and choose your own app, things immediately become a bit complicated.

Slightly technical as it may be, it’s also very promising. The early adopters stand ready to help, all the features (and more) that you may want from a Twitter/X-like platform are available and the non-algorithmic focus imparts a feeling of control. On Mastodon you may be slightly confused, but you’re also very much in charge.

Of course, what you’re not getting is content going viral to a massive audience. Mastodon is thinly populated at this time. So we’re decidedly not there for the reach but because we see clear affinities with our openness ambitions, because the platform’s open architecture may allow for really interesting re-use/automatisation and because there might be a time where the current social media behemoths lose steam. In which case our mastodon riding skills may well come in handy.

We’re starting small. But we see great potential – not least for joining forces across museums and other fine cultural institutions. French cultural institutions are getting together at ReseauCulture.fr – and perhaps Danish/Nordic ones should look very closely at that model. Hit us up if you’d like to talk! 🤗

🏠 SMK on Mastodon (we speak Danish)

Our first Mastodon post

Je n’ai jamais connu la guerre. Je suis issu d’une génération privilégiée. Né au bon moment. Né au bon endroit.

Mes parents l’ont connue.

Mon père, né en 1939, ne pouvait pas savoir, au moment de fêter son cinquième anniversaire, ce que c’était que la paix.

Ma mère, née en 1944, a bien failli ne pas être ma mère. Ses frères et sœurs, dans leur précipitation, l’ont une nuit oubliée en courant se réfugier dans la cave. Les bombes ne sont pas tombées loin, une armoire a basculé sur son berceau. Ils étaient solides les berceaux, en ce temps là.

Comme tant d’autres de sa génération, mon père est allé en Algérie. On ne lui a pas demandé son avis… Il en est revenu, lui.

Je suis né quelques années après.

Mes grand parents ont connu deux guerres. Mes arrières grands parents ont connu deux guerres. Je pourrais continuer longtemps comme ça, si je le voulais.

Je l’ai dit : génération privilégiée.

Et pourtant…

Même si je n’ai pas connu la guerre, même si je ne l’ai pas vécue dans ma chair, ce que j’en sait me suffit largement.

Mais la guerre n’est pas un choix.

On fait la guerre- par défaut – parce qu’il est trop tard – parce que les autres options sont épuisées – parce que les décisions qu’il aurait fallu prendre à temps n’ont pas été prises – parce qu’à un moment le courage à manqué.

Parce que les puissants ont manqué de courage. Parce qu’ils n’ont pas su écouter. Parce qu’ils croyaient savoir. Parce qu’ils étaient bien trop arrogants pour reconnaître qu’ils s’étaient trompés.

Alors ils en envoient d’autres se faire tuer.

C’est compliqué la paix.

Ça demande du courage, de l’écoute, du respect.

Alors que c’est si simple la guerre.

Il y a les gentils, il y a les mauvais.

Et puis, il y a tant d’argent à gagner…

^{Chapter Index ARFGAM contains mature themes and at times is #NSFW}

After leaving Michelle, Stelmaria and Chad were having a conversation about whether she should make herself known to his family. Surprisingly, Chad's position was the logical one. Unsurprisingly, Stelmaria would convince him to agree to her position. This conversation was long, repetitive, and boring. Luckily, more interesting things were happing on ethereal side of reality.

***

Sebastian whistled at his workbench and chipped away at the piece of petrified wood. Making petrified wood chips was the one simple alchemical task he had never assigned to an assistant. He found it relaxing. His boss Trinity had a long standing complaint that it was 'a simple task way below his pay grade'. Since Sebastian knew all she would ever do about it is tell him in performance reviews that his refusal to allocate this task was 'beyond unacceptable and incredibly petty,' so he cheerfully ignored her complaints.

Sebastian had been working as an alchemist at the Ethereal Alliance Tranquility Council's Office of Courageous, Kind, & Laudable Interdisciplinary Civilian Knights that Virtuously and Ardently Guarantee the Terrific Omnipresent Civilized Utopia is Maintained for decades. Growing up Sebastian had been sure of two things. The first was that he wanted to be an alchemist. The second was that his family was too poor to afford the admission fees for an alchemical school. As a result he pursued an Alliance program that would cover his education in exchange for ten years of government service following its completion. When he first learned he had been assigned to the Knights, he assumed the ten years of service would be miserable. By year five he had come to realize that he loved being a knight.

Three years after this realization that he met his spouse, Taylor. They were an heir to a sizable family fortune dating back to pre-Alliance times. When they proposed to him it made the decision on if he should leave the Knights for a higher paying job when his required service was complete, an easy one.

Now at the age of sixty-seven, he found himself as both the Chief of Alchemy and Assistant Director, Field Force Support. The latter of which he had become because fifteen years prior when the AD2FS who had been his primary mentor retired, Sebastian informed Giovanni, the Knight Director, he did not want to report to “whoever he thinks can do the job justice”. Giovanni, agreed that Sebastian could report directly to him. Sebastian loved this arrangement because the Knight Director has l zero time to spend managing the Chief of Alchemy.

Eight years later Giovanni retired. The current Knight Director, Trinity, succeeded him. Unfortunately for Sebastian, Trinity happened to be the AD2FS that he had refused to report to. The first thing Trinity said to Sebastian after being named Knight Director was “I have an amazing opportunity for you.”

Sebastian hated the administrative bullshit that went with being AD2FS, but cared about the Knight's mission enough to do a good chunk of it. He did all the stuff he thought was actually necessary for the success of the mission. He skipped all the dumb bureaucratic stuff like attending departmental budget reviews or Trinity's weekly leadership meetings.

Sebastian had been making petrified wood chips for the last six hours. It was one of his days off. He was stressed. If the mission had gone remotely well, Des would have been back yesterday. At thirty and three Despoina was the youngest Field Knight in the twelve millennia of the Knight's existence to be named Regnar-Leas, the highest field rank. In Sebastian's opinion, she was on her way to being the greatest Knight the Alliance would ever see. Still, he disagreed with her and Trinity's decision to send her on this mission solo. The delay in her return was all the proof he needed that he was right.

Chipping the petrified wood was the only thing keeping him from imagining how he was going to explain to Taylor that a Massena Escapee brutally murdered their little sister and was most likely harvesting souls in Tempo to try and grow in power because she had been raised in a cult dedicated to reestablishing the ancient Seelie Court and murdering all fiends as well as any älva and kodoma subtypes traditional folklore tends to associates with the Unseelie. It was not a conversation he wanted to have with his spouse.

The chunk he was working on had been as big as his head at the start of the day, it was now smaller than his thumb. He did not have any more, and he knew it. His hearts were an anxious tick-thunk, tick-thunk, like a pair of drums in his chest. When the steady beat was interrupted by the screech of his office door swinging open, he was so surprised he nearly fell off his stool.

The hand that smacked a pile of silver coins down on his desk was covered in dry blood and mud, but there was purple paint on its long pointy finger-nails. Sebastian felt every one of his muscles relax. There was one more tick-thunk and then he only heard his sister-in laws heavy breathing. He spun on his stool and jumped into her arms.

“What you are not going to gloat?” Des managed to say, she sounded exhausted.

Gloat? Sebastian was not sure what she could possibly be talking about. “I have been chipping petrified wood for hours Des. I thought I was going to be telling Tee about your brutal execution by a lunatic. I would have made Trinity tell your parents though.”

“Well now I am a bit disappointed. I guess surviving means that I lost my bet to you and that Trinity gets to avoid a conversation she would have hated.”

“What did the boss say when you gave her your report?”

“Errrrr, I do not know. I have not seen her yet. I just got back.”

“Despoina, why in the Arcanum did you not send a report ahead? This was critical-CRITICAL shit. I told you time and again the samples I looked at suggested that Stelmaria may be the strongest hyōsei to exist in this age. Trinity has probably been preparing a report for the Tranquility Council since yesterday morning assuming...” Sebastian chuckled and leaned onto his young sister in-law. “Fuck Des, even I would not dare to fuck with the boss that much. Does it make me a bad AD that I am proud? Rhetorical question, I do not care if I am a bad AD that would be Trinity's fault for being petty and promoting me.”

Sebastian looked up at Despoina's face for the first time. There was not a hint of amusement in her face which was strange because fucking with Trinity was pretty much the only thing that made his sister in-law smile. “You are being earily quiet Des, keep it up and I will call you 'Little Si...Ouch”

“I may be about to collapse you shithead but I am not deaf. Nobody calls me that but the one sibling I still talk to. That includes that one sibling's husband who I regrettably have to interact with as part of my job. I am juiced. Alchemical. Blood. Right. Now. Before. I. Decide. To. Drink. Yours.”

Sebastian rolled his eyes at that. “Des I know, you know that gnome magic cannot be harvested even by a pterafri sucking our blood. I also know you do not even eat fish because you do not want to be like your parents. Why do you need Alchemical? You will have enough time to spend sleeping with a bloodstone before Trinity assigns you something knew. The side effects can be pretty bad if it is not made right you know?”

“Seb, I did not get her. She escaped...with a human.”

“Are you fucking with me?”

Despoina covered her face with her hands and shook her head. “He shot me, in the eye, with an iron tipped wooden arrow.”

“GAHAHAHAAHAAAHAAAAA,” Sebastian fell to the floor laughing. He knew it was wrong but a puny human besting his sister in-law with a human made weapon was the most ridiculous thing he had ever heard. “Sorry, sorry, sorry I know this is serious this is just...what are the odds? Arcane shit. It had to be pure iron Des its the only way right? Fuck that is impossible and then the shaft was wood? Fuck, do you know most human weapons have not been made out of pure iron in centuries?” Sebastian, took a deep breath. “Sorry, I am sorry this is serious.”

He got up and grabbed a key ring out of his pocket and unlocked a small silver chest sitting on his work bench. “I made some yesterday afternoon when you had not returned. Just in case. There are exactly three fairies in the known realm that could make a better substitute for the real thing. I am not saying there will not be side effects but I am saying I feel pretty confident none of them will be permanent.” He held up a large glass phial about half the size of a bottle of wine. It was filled with a thick maroon cream.

Despoina snatched the phial out of his hand, popped open the lid and sniffed it. “How many times have you made Alchemical Blood Sebastian?”

Sebastian watched sweat pool on Des' brow as she examined his work. “Yesterday was the first, second, third, and fourth times I have made Alchemical blood. I destroyed the first two due to lack of color uniformity, which some studies have linked to more extreme crashes. The third I feared was rather low in potency based on arcane readings so I destroyed that one after I made this one, which is the fourth. As I said there might be a handful of fairies that could do better.”

Sebastian crossed his arms and smirked as he added, “It costs a small fortune to make a single phial, the entire annual alchemical supply budget the council gives me would have been wiped out by what it cost me to make these four. I bought the ingredients with your family's money. I told Taylor you had a mission that was dangerous enough I wanted to have some special ingredients outside of the Council's budget just in case you came back with something that would otherwise be very painful to treat.”

Des grimaced, “did they buy that?”

“Nope. They said to tell you that craving blood is a part of who you are and does not make you your mother.”

“I am going to gut that fucking fur ball for making me do this,” Despoina vowed. Then she tilted her head back, lifted the phial to her lips, and slammed it back in three giant gulps. It was sickly sweet. “Not as disgusting as your mint flavored memory salve, but a close second.”

She turned and walked out the door. Magically mending and cleaning her soiled clothes with a wave of her hand on her way out.

***

Meanwhile back in the temporal side of reality Stelmaria had eventually gotten Chad to agree with her by offering him a boob job. He was now passed out on his bed snoring loudly. Stelmaria said a quick chant to make sure her cute bean would stay asleep while she was out hunting. Then jumped out his bedroom window, shifted into her beast form, and trotted off towards town.

AUTHOR's NOTE: The reader is likely to notice that as we explored the Ethereal Alliance's civilization that the there are abbreviations, acronyms, and slang was common. Some of these readers may be thinking something akin to 'So I am just supposed to accept that this fantasy world is based in the English language and Latin alphabet or that somehow their wordplay can translates seamlessly to ours in a way that sometimes creates puns in English by coincidence?' For these readers I would like to offer two different explanatory responses for them to choose from.

OPTION A: Of course not, the language and alphabetS in the world our story is set in in more fantastical and complex than most could comprehend (but not you because you are so brilliant and smart). There writing system was a combination of logographic, phonetic, syllabic, and enigmatic characters (some pictorial and some abstract). The cleanness of the abbreviations has only been added here to make reading easier for less intelligent readers. If this is breaking your immersion I apologize and if there is ever a super deluxe sprayed edged limited premium edition collector's copy, rest assured I pay to have it written in an original language that only people as smart as you can understand.

OPTION B – “Are you aware that most of the worlds languages use acronyms, abbreviations, and slang? That even some of the ancient languages used them? The ancient Greeks and Ancient Romans used both acronyms and abbreviations. I looked it up and there are examples of abbreviations in ancient Mayan written language as well (which I only selected as my example because my (high school level) understanding of ancient history is the Mayans were isolated from Eurasia and Africa until the sixteenth century). I am not an expert on classic literature (also high school level understanding) but I think there are thgere puns in Beowulf, the Illiad/Odyssey, and One Thousand and One Nights? One thing I know for sure is both testamets of the christian bible have wordplay. What was the point of this author's note? No you are not supposed to even think about that so you should not have to accept it. As stated in the forward this story is poorly written. I sincerely apologize if this is a pet peeve of yours and my decision to do this has mad you sad. Let me try to move your mood in a different direction. Despite spending winters alone in the middle of the ocean, Atlantic Puffins are monogamous and mate for life. They come back to the same nest with the same partner year after year. Its like an annual second chance romance or lost lovers or something. How fun!”

< Chapter 4

Main Blog

Chapter Index

^{#Romantasy #RomantasyforMen #Satire}

9 March 2026

Face shield bag (working title): was walking in Vauxhall and found the outer packaging for a set of CPR mannequin shields. Made of transparent plastic, on which was printed a wonderfully-poorly-rendered line drawing diagram showing how to use the product—hands affixing a shield to a mannequin's lifeless face, another (living?) face entering the diagram's second stage to put its lips to the first one. All folded in on itself and resting delicately over sparse weeds sprouting from wet soil squeezed up against a concrete curb. Something about it brought to mind Polke's watchtower series (particularly Watchtower (Hochsitz) from 1984), both in mood—relaxed at a kind of equilibrium but sinister—and visual complexity—the bent plastic packaging caught daylight at odd angles, blocking visibility of the weeds, soil, and diagram here and there. What resulted is a painting that to me feels ancient, like a hieroglyph partially lost to material decay. Which sits in an odd harmony with the satisfaction on the face floating at the top of the composition. The color is indebted to Eliot Porter's Winter Wren, Great Spruce Head Island, Maine (1960), which holds an aspirational kind of long-ago-now-ness that I'm permanently searching for.

#books #nonfiction #economics

“There is no magic money tree” is the stern injunction invoked by politicians, central bankers and economists to explain to a fiscally imprudent public why it cannot have nice things. Fiscal rectitude is now the primary virtue of government, perhaps nowhere more so than in the United Kingdom where the Treasury has shackled itself to the need for approval from an ‘Office for Budget Responsibility’. Running deficits or printing money, we are told, is only one tiny step away from Weimar Republic levels of financial calamity.

But what if it wasn’t thus? That is the alluring promise of Modern Monetary Theory (MMT), which first gained prominence in the wake of the Great Recession and argues that not only can governments print money to cover expenses, but they should do so to fully realise a nation’s productive capacity. It is a provocative and controversial theory that repudiates the need for permanent austerity in the name of balanced budgets, and finds one of its most ardent advocates in Stephanie Kelton, erstwhile chief economist to US senator Bernie Sanders. In her book The Deficit Myth, she takes her argument for a ‘people’s economy’ built on the insights of MMT to a wider audience.

The Deficit Myth faces the triple challenge of any non-fiction book that assails an existing orthodoxy. It must set out a compelling argument, be intelligible to a lay audience, and dispel hegemonic common sense. This is a daunting task, and the meagre evidence base, weaknesses in Kelton’s writing style, and a different perspective on political economy meant I was left unpersuaded by the book’s stronger claims. Nonetheless, it is a thought-provoking read that provides ample critique of economic orthodoxy, and left me receptive to exploring more rigorous defences of MMT in future.

MMT is based on the chartalist premise that any government operating a freely floating fiat currency can create money to fund its expenditure. Governments can do this because they are the monopoly issuer of their own currency, which means they can never ‘run out’ of it. This money then flows into the economy, from which it can be removed through taxation, to avoid an excess supply of money driving up inflation. I had come across this view before in David Graeber’s Debt: The first 5000 years, although Graeber did not explore the economic consequences as thoroughly as Kelton.

MMT pre-empts the obvious counterargument that governments creating money will simply lead to inflation by positing that the additional money will be absorbed by unutilised productive capacity in the economy instead. In essence, MMT argues that it is not the relationship between money and goods and services available to purchase that drives inflation, but the relationship of money to the productive capacity of an economy to produce goods and services. As long as increased government expenditure results in a commensurate increase in things available for purchase, by stimulating economic activity, inflation will be kept at bay. It is only when the economy is running at full capacity that increasing the money supply further will cause inflation.

This argument will sound familiar to readers acquainted with (post-)Keynesian theories advocating countercyclical spending to dampen the negative effects of the business cycle, and my impression is that different views on the nature of money notwithstanding, MMT advocates and Keynesian economists are at least fellow travelers. It is an intriguing and logically coherent hypothesis, but unfortunately The Deficit Myth does not offer much evidence to buttress the initial premise. Kelton references a number of other heterodox economists, but unless the reader is already predisposed to agree with the argument, the appeal to authority does not work if the reader is unfamiliar with the sources cited, yet still aware that they are not universally accepted.

This lack of evidence and a failure to address any of the obvious critiques or counterexamples to MMT leave the central argument in a precariously weak position after the first two chapters, and The Deficit Myth does little to shore it up in the remainder of the book. Instead, it applies the core premise to a range of policy issues, such as the national debt, trade imbalances and social security commitments. The Deficit Myth’s prescriptions follow logically from the core premise, which Kelton repeats somewhat overmuch, but they do not offer further proof for its truthfulness. If one does not accept MMT’s core tenets, the whole argument immediately falls apart.

Avoiding substantive engagement with critiques of MMT also leads Kelton into a dead end when explaining why her theory isn’t universally accepted. Discounting competing views on MMT’s validity, she resorts instead to ascribing the failure of policymakers and mainstream economists to accept MMT to either an almost delusional psychological investment in the myth that government finances work similarly to a household budget, or to bad faith ploys for continued austerity. I certainly don’t dispute that hegemonic dogma constrains how people think, but it is not persuasive as the only reason why so many economists, including from heterodox traditions, remain stubbornly unconvinced of MMT’s validity.

All this leaves Kelton’s account of MMT exposed to numerous lines of attack. There is no account of how MMT would explain or manage crises such as the 1970s stagflation or the hyperinflation seen in Weimar Germany or contemporary Zimbabwe or Iran. The chapter proposing a jobs guarantee, does not work through how a strengthened bargaining position for labour might cascade through the economy. The chapter on international trade explains the dangers of governments restricting their monetary sovereignty by linking their own currency to that of another country (usually the dollar), but does not address the risk of increases in the money supply or trade deficits causing currency devaluation, making imports more expensive. This chapter also suffers most from the US-centric perspective in The Deficit Myth, because while Kelton notes the exceptional position of the United States as the issuer of the world reserve currency, there is little exploration of the advantages this confers on the US, and the disadvantages it poses for everyone else. No other country can rely on a near-infinite demand for its own currency to maintain favourable exchange rates despite running structural trade deficits. The lack of consideration for higher-order effects certainly makes the book more readable, but on the flipside also makes it feel so too simplistic to remain persuasive.

The last couple of chapters are dedicated to policy problems that Kelton argues are far more important ‘deficits’, such as crumbling infrastructure, inequality, the atrocious healthcare provision in the United States, and imminent environmental collapse. It is these chapters where The Deficit Myth cannot cash the cheques it wrote for itself at the start of the book. Kelton reminds us that it is the productive capacity of the real economy, not the supply of money, that is the real constraint on what is achievable. Yet as we near the end of the book, we are no wiser on what this capacity is, how we would know what it is, and how it is constituted. It is plausible that MMT could solve any of the problems identified by Kelton individually, but it is doubtful it can solve them all at the same time. The Deficit Myth offers no evidence that simply increasing the money supply would enable us to pay for better healthcare and environmental restoration and a jobs guarantee and infrastructure repair and any of the other things Kelton cares about. Instead, Kelton has to concede that other policy measures, such as progressive taxation, environmental legislation, universal healthcare and industrial policy will also be required.

And so we find ourselves back at the political in political economy. The allure of MMT is its promise of a technical fix to a political problem, and Kelton repeatedly stresses that MMT is not ideology but monetary reality. But in the end, until we achieve fully automated luxury communism, we cannot escape political struggle over our societies’ limited productive forces. Kelton falls into the same trap as Rutger Bregman in Utopia for Realists by proposing an ostensibly objectively positive policy as a shortcut to avoid class conflict, but with class antagonism itself standing in the way of the policy being implemented. As Cory Doctorow reminds us, if something is good for workers, the bosses will hate it. The reason why we cannot have nice things is not because of a mismatch between the money supply and productive forces, but because it is not in the interest of the capitalist class to let us have them. Universal healthcare and a jobs guarantee may well benefit society in the abstract, but the bosses know that insecure, desperate workers are much easier to discipline and exploit.

Jane McAlevey said it best. There are no shortcuts. MMT may well be a useful tool in the hands of labour, but if so, it will still require a powerful working class to wield it.

In the end, I was not persuaded that MMT is the gamechanger that Kelton propounds it to be. But The Deficit Myth remains a valuable and critical intervention in public debate on how we run the economy, and a powerful argument for removing artificial constraints on our welfare and prosperity. The book could be read as a first step on one’s MMT journey, rather than the final word, and I expect more in-depth MMT works would address the critiques I’ve raised. As we discovered to our collective detriment following the Great Recession, there are dangers in having a monoculture of economic theory. It will however take more robust defences and sharper arguments if Kelton wants to see MMT emerge victorious.

Notes & Suggestions

• In her early chapters, Kelton introduces the concept of the ‘non-accelerating inflation rate of unemployment’ (NAIRU) which is the putative reference point for central banks to manage inflation. It isn’t central to Kelton’s argument and so I didn’t manage to integrate it into the blog proper, but it was the part of the book that made me most furious. The NAIRU is a rate of ‘acceptable’ unemployment that unelected technocrats decide is necessary to avoid inflation. In other words, it is official government policy in all advanced economies to keep a certain percentage of the workforce out of work as a reserve army of labour. Yet the same politicians and technocrats who deliberately inflict suffering and misery on millions of workers for the benefit of ‘the economy’ will simultaneously blame the unemployed for their situation. If there is one other thing to take away from The Deficit Myth, it is that we should never again take serious any politician talking about ‘skivers vs strivers’ or ‘encouraging’ the unemployed back to work.
• Kelton’s book is a decent entry point, but there is a lot more on MMT out there, on both sides of the debate. Jacobin has featured an extensive critique and rebuttal, as well as an interview with Stephanie Kelton. In the UK, Richard Murphy is one of the more vocal MMT advocates.

Bal Bhavan Mayur Vihar Hosts CBSE Central Zone Girls’ Football with Grandeur, Spirit, and Inspiration

visite : https://www.educationtoday.co/magazines/north-issue-2025/bal-bhavan-mayur-vihar-hosts-cbse-central-zone-girls-football-with-grandeur-spirit-and-inspiration

Bal Bhavan Public School, Mayur Vihar Phase-II, once again proved why it stands as a beacon of holistic education and student empowerment in Delhi. The school hosted the prestigious CBSE Central Zone Girls’ Football Tournament (U-14, U-17, U-19 categories), transforming the Sh. G.C. Lagan Memorial Sports Stadium is transformed into a lively arena of passion, sportsmanship, and resilience. Over nine days, the event brought together 1,000 young athletes from leading schools across Delhi, creating an inspiring platform that went beyond competition and embodied the spirit of NEP 2020 Sports education. For Bal Bhavan, the event was not just about football; it celebrated girls’ empowerment through sports, reinforcing values like teamwork, discipline, and confidence. It was also a testimony to the school’s unwavering vision of nurturing well-rounded individuals. A Grand Inauguration with Visionary Leaders The tournament commenced with an opening ceremony graced by eminent dignitaries. IAS Veditha Reddy, Director Education and Sports, Govt. of NCT—Delhi, attended as the Chief Guest. Her presence was symbolic of the government’s commitment to promoting school football in Delhi and creating equal opportunities for young girls in sports. Speaking at the event, she highlighted how sports catalyze confidence and leadership in young learners, particularly girls. She emphasized that the lessons learned on the football field—discipline, courage, and resilience—are the same principles that shape successful leaders of tomorrow. The Guest of Honour, Shri Jagseer Singh, an Arjuna Awardee and Paralympic athlete, left the audience deeply moved with his life journey. He reminded the athletes that while trophies and medals are temporary, the courage to face challenges and the willpower to persevere define true champions. His words resonated across the stadium, inspiring every participant to give their best, regardless of the outcome. A Gathering of Strong Pillars of Education The ceremony also saw the presence of the distinguished leadership team of Bal Bhavan Public School and respected educationists, whose contributions have shaped the school into one of Delhi’s most admired institutions. Among them were: Shri Ramesh Kandpal Ji, Senior Consultant of Akhil Bharatiya Anuvrat Nyas Shri B.B. Gupta Ji, Director of the Institution Ms. Rachna Gupta, Former Principal of Bal Bhavan Swasthya Vihar Branch Ms. Rinni Srivastava, Management Committee Member Shri G.S. Grover Ji, Chairman of the School Ms. Neha Gupta, Principal of Bal Bhavan Junior Branch Mr. Vividh Gupta, Principal of Bal Bhavan Public School Ms. Kavita Mehrotra, School Manager Their collective presence symbolized the school’s strong commitment to holistic growth. As Principal Mr. Vividh Gupta remarked, the essence of the tournament lay in “fair play, integrity, and the holistic growth of every learner.” His words reflected Bal Bhavan’s belief in fostering 21st-century skills, global citizenship, and strong values through academics and sports. Ceremonial Splendor: March Past, Oath, and Unity The event began with a traditional lamp-lighting ceremony, followed by a grand March Past, filling the stadium with vibrant colors, synchronized steps, and the spirit of unity. Athletes representing various schools marched proudly, their enthusiasm embodying the motto of sports as a unifying force. The oath-taking ceremony further reinforced values of integrity, respect, and resilience. Principal Mr. Vividh Gupta, addressing the athletes, spoke of the importance of fair play and the school’s role in cultivating achievers and responsible global citizens. His commitment to nurturing learners who embody both knowledge and values was evident in his words. Inspirational Messages that Resonate IAS Veditha Reddy urged young girls to view sports as a game and a foundation for self-belief and leadership. She pointed out that participation in tournaments like these reflects the vision of NEP 2020 Sports education, which advocates for equal opportunities, physical well-being, and character development in every student. In his motivational address, Jagseer Singh shared glimpses of his journey as a Paralympic athlete. His life, marked by challenges and victories, became a lesson of courage and dedication. He encouraged the young footballers to embrace setbacks as stepping stones and to chase excellence with perseverance. His message, “Trophies may belong to one, but the courage to play belongs to all,” echoed through the stadium, leaving a profound impact. A Special Gesture of Blessings In a symbolic moment, the Chief Guest and Guest of Honour signed a commemorative mug, a gesture that will find a permanent place on the school’s Shelf of Memories. This token marked their presence and became a source of inspiration for future athletes, reminding them that their efforts are recognized and celebrated. Kick-off and Competition Spirit The tournament officially began with a ceremonial toss and handshake between Bal Bhavan Public School and Army Public School Dhaula Kuan. As the ball rolled onto the field, students, teachers, and supporters cheered. Over the next nine days, the stadium would witness matches filled with energy, strategy, and resilience. For Bal Bhavan Public School, hosting the CBSE Football Tournament was about much more than the scoreboard. It was about teaching students the value of teamwork, discipline, and sportsmanship—qualities that extend far beyond the football field into every sphere of life. Sports as a Pathway to Empowerment The CBSE Central Zone Girls’ Football Tournament at Bal Bhavan, Mayur Vihar, powerfully reflected girls’ empowerment through sports. Each match reminded us that education is not confined to classrooms; it is equally about experiences that shape courage, character, and resilience. By aligning the tournament with the vision of NEP 2020 Sports, the school highlighted its progressive approach to holistic learning. Girls taking to the football field in large numbers symbolized a cultural shift where sports are becoming a pathway for confidence, leadership, and equal opportunities in society. A Legacy of Inspiration As the nine-day tournament unfolds, Bal Bhavan Public School celebrates victories and the spirit of participation. The school firmly believes that while trophies belong to one, the lessons, friendships, and values learned belong to all. This legacy of inspiration reflects the school’s mission to nurture confident, compassionate, and equipped learners with the skills required for the 21st century. Through events like these, Bal Bhavan Mayur Vihar Sports initiatives set benchmarks for how schools can empower young girls, promote inclusivity, and contribute meaningfully to the nation’s future. Conclusion The CBSE Central Zone Girls’ Football Tournament by Bal Bhavan Public School, Mayur Vihar, was more than a sports event; it was a landmark in reinforcing the values of resilience, empowerment, and unity. Guided by visionary leaders like IAS Veditha Reddy and inspired by icons like Jagseer Singh, the event left a lasting impression on every participant and spectator. For Delhi and beyond, it is a reminder that school football is not just about matches but about building confident individuals who will lead society with strength and vision. In the words of Principal Mr. Vividh Gupta, the school will continue to “celebrate the passion, dedication, and spirit of every young athlete,” ensuring that the courage to play and the lessons learned remain with students for a lifetime.

Toilet Rol Stemmingsmeter van de Target Marketeer

Hallo Toilet Klant fijn dat u weer bent ingelogd voor deelname aan de persoonlijke stemmingsmeting in het mini eerste kamertje.

Dit door de enige echte toiletpapier influencer aangemaakte vul evenement kost u maar een paar minuten.

De maatschappij heeft mij geroepen zodat ik hier elke dag uw stemming mag meten. Zodat zij van de maatschap dankzij mij weten wat er op dit moment in u omgaat. Deze verse gegevens zetten zij later in voor het aanbieden van unieke massaal gefabriceerde producten voor in u geest of op locaties rondom het lijf. Dat wilt u toch ook!

Nu de test te vinden op vel twee, drie en vier, opeenvolgend, in drievoud hangend aan de rol.

Vraag 1.

Hoe voelt u zich op dit moment?

1. Enorm blij
2. Opgetogen
3. Gewoon
4. Niet jedat
5. In mineur

Veeg nu het juiste aantal keer over uwer edele billen en deponeer het antwoord in de pot.

Vraag 2.

Wat gaat u straks doen?

1. Genieten van de hele wereld
2. Allemaal nieuwe spullen aanschaffen
3. Wat ik altijd doe
4. Alleen de dingen die ik moet doen
5. Op bed liggen en heel lang stil liggen huilen

Veeg nu het juiste aantal keer over uwer edele billen en deponeer het antwoord in de pot.

Vraag 3.

Bent u tevreden over het vandaag reeds behaalde?

1. Uitermate tevreden alles ging beter dan goed
2. Niet ontevreden er is aardig wat gelukt.
3. Gewoon, zoals het hoort te gaan ging het
4. Niet echt tevreden, er ging meer fout dan goed
5. Diep ongelukkig, de paar dingen die ik deed zijn volkomen mislukt

Veeg nu het juiste aantal keer over uwer edele billen en deponeer het antwoord in de pot.

Spoel na afloop de aangeveegde test door naar de daarop wachtende instanties. Aan het einde van de dag kunt u online de dag aanbiedingen zien die horen bij de uitslag van de wc rol stemmings test van vandaag. Ik dank u wederom voor u inzet en tot straks.

In today’s fast-paced digital world, businesses need smarter ways to interact with customers, provide instant support, and improve operational efficiency. Artificial Intelligence (AI) and chatbot solutions have emerged as powerful technologies that help businesses automate communication, enhance customer experiences, and streamline processes. From answering customer queries to generating leads and improving service availability, AI-powered chatbots are changing how companies connect with their audiences.

Understanding AI and Chatbot Solutions

AI refers to the use of intelligent systems that can analyze data, learn patterns, and make decisions with minimal human intervention. Chatbots are one of the most practical applications of AI in business communication. They are automated programs designed to simulate human conversation and assist users through text or voice interactions.

AI chatbots can be integrated into websites, mobile apps, and messaging platforms to provide instant responses to customer queries. Unlike traditional customer support systems that rely on human agents, chatbots can handle multiple conversations simultaneously, ensuring that users receive quick and consistent responses at any time.

Why Businesses Are Adopting AI Chatbots

Modern customers expect fast and convenient communication when interacting with a brand. Waiting for long response times can lead to frustration and lost opportunities. AI chatbots help solve this problem by providing real-time assistance.

Businesses are adopting chatbot solutions because they help:

Provide 24/7 customer support

Respond instantly to customer inquiries

Improve user engagement on websites

Reduce operational costs

Capture and qualify leads automatically

By automating repetitive tasks and frequently asked questions, companies can allow their teams to focus on more complex and strategic work.

Enhancing Customer Experience

One of the biggest advantages of AI chatbots is their ability to enhance the overall customer experience. When visitors land on a website, they often look for quick answers regarding products, services, pricing, or support. A chatbot can immediately assist them, guiding them through the website and providing relevant information.

This level of instant interaction creates a smoother user journey and increases the chances of converting visitors into customers.

AI Chatbots for Lead Generation

AI chatbots are also highly effective tools for lead generation. By engaging visitors in conversation, chatbots can collect valuable information such as names, email addresses, and business requirements.

For example, when a visitor asks about a service, the chatbot can respond with helpful information and then request contact details for further assistance. This automated process helps businesses capture leads even outside working hours.

Automation and Business Efficiency

Automation is one of the key reasons businesses invest in AI solutions. Chatbots can automate tasks like answering common questions, scheduling appointments, sharing product details, and providing customer support updates.

This automation improves efficiency and reduces the workload for customer support teams. As a result, businesses can operate more efficiently while maintaining high levels of customer satisfaction.

The Future of AI-Powered Communication

As technology continues to evolve, AI chatbots are becoming more intelligent and capable of understanding natural language and complex queries. Advanced chatbot systems can analyze user behavior, provide personalized responses, and integrate with CRM systems and marketing platforms.

In the future, AI-driven communication will become an essential part of every digital strategy, helping businesses deliver faster, smarter, and more personalized experiences.

AI & Chatbot Solutions by Skyno Digital

At Skyno Digital, AI and chatbot solutions are designed to help businesses automate communication, improve customer engagement, and increase lead generation. By combining advanced AI technology with user-focused design, Skyno Digital builds intelligent chatbot systems that integrate seamlessly with websites and digital platforms.

These solutions enable businesses to provide instant support, capture valuable customer insights, and improve operational efficiency while delivering a better digital experience.

Conclusion

AI and chatbot solutions are transforming the way businesses communicate with customers. By providing instant responses, automating routine tasks, and improving user engagement, chatbots help companies operate more efficiently and deliver better service.

For businesses looking to stay competitive in the digital era, implementing AI-powered chatbot solutions can be a smart step toward building stronger customer relationships and achieving sustainable growth.

Freundinnen & Freunde der Weisheit! All die Selbstoptimierungsgurus erklären Glück zur reinen Privatsache. Echtes Glück ist aber etwas anderes: nicht eine fortlaufende Selbstinszenierung, sondern Verantwortung und Sinn.

Unsere gegenwärtige Vorstellung von Glück ist oft erstaunlich schmal geraten. Was früher mit Tugend, Gemeinsinn und einer gerechten Gesellschaft verknüpft war, erscheint heute als Frage individueller Befindlichkeit: Wie kann ich mich besser fühlen, produktiver werden, meine Ziele effizienter erreichen? In einer Kultur, die Selbstoptimierung, Wohlfühlroutinen und persönliche Markenbildung zum Ideal erhoben hat, ist das Verständnis von Glück zur Privatsache geworden – reduziert auf Momente der Zufriedenheit und messbar in Datenpunkten. Was dabei verloren geht, ist die tiefere Dimension von Glück: jene, die sich aus Zugehörigkeit, Verantwortung und Sinn ergibt.

Ein erster Schritt zu einem tragfähigeren Glück liegt in der Abkehr vom rein individuellen Fokus. Wer sich als Teil eines grösseren Zusammenhangs begreift – sei es im Familienkreis, im Gemeinwesen oder in einer freiwilligen Aufgabe –, erlebt sein Leben nicht nur als fortlaufende Selbstinszenierung, sondern als bedeutungsvoll durch Verbindung. Gerade in einer Zeit, in der viele gesellschaftliche Strukturen unter Druck stehen, gewinnt das bewusste „Für-andere-da-sein“ an Wert – nicht als moralische Pflicht, sondern als Quelle innerer Stimmigkeit. Es sind oft die kleinen, unsichtbaren Beiträge, die Beziehungen tragen und persönliche Erfüllung ermöglichen.

Zweitens lohnt sich ein Perspektivenwechsel: Glück ist nicht primär eine Frage des Konsums oder der Wahlfreiheit, sondern eine Frage der Ausrichtung. Studien zeigen, dass Menschen ihr Leben dann als sinnvoll erleben, wenn sie ihre Handlungen mit grösseren Werten verbinden – etwa Fürsorge, Gerechtigkeit oder Verlässlichkeit. Es braucht kein perfektes Leben mit durchgeplantem Alltag. Viel entscheidender ist, ob wir unser Handeln als kohärent und relevant empfinden. Wer sein Engagement ausrichtet auf etwas, das über das eigene Wohlbefinden hinausgeht, erfährt oft eine tiefere Form von Zufriedenheit.

Drittens sollten wir die gängige Vorstellung hinterfragen, dass Glück mit ständiger positiver Stimmung gleichzusetzen sei. Ein erfülltes Leben schliesst Ambivalenz, Anstrengung und Unsicherheit mit ein. Gerade in Zeiten von sozialer oder ökologischer Krise zeigt sich, dass Glück nicht im Rückzug liegt, sondern im aktiven Mitgestalten einer Welt, die für viele lebenswert bleibt. Das grosse Glück entsteht dort, wo Menschen Verantwortung übernehmen, ohne sich aufzuspielen – wo sie verlässlich handeln, ohne immer perfekt sein zu müssen. Glück ist dann nicht das Ziel, sondern das Echo eines geglückten Daseins.

Denkanstoss zum Wochenbeginn

„Ich kann ihnen nicht sagen, wie man schnell reich wird. Ich kann ihnen aber sagen, wie man schnell arm wird: indem man nämlich versucht, schnell reich zu werden.“ – André Kostolany (1906–1999)

ProductivityPorn-Tipp der Woche: Schwierige Aufgaben zuerst (Eat the Frog)

Fange den Tag mit der schwierigsten oder unangenehmsten Aufgabe an. Danach fühlt sich alles andere leichter an und du vermeidest das ständige Aufschieben.

Aus dem Archiv: Warum nicht jeder Ratschlag zu dir passt

Wir hören es immer wieder: Erfolgreiche CEOs sagen, dass das Geheimnis ihres Erfolgs darin liegt, „Nein“ zu sagen. Influencer raten uns, dieses oder jenes Produkt zu kaufen, weil sie es angeblich selbst lieben. Doch was für sie funktioniert, muss nicht automatisch für dich passen. Pauschale Ratschläge ohne Berücksichtigung deines eigenen Kontextes können sogar gefährlich sein. Dieser Beitrag wirft einen kritischen Blick darauf, warum es so wichtig ist, Ratschläge zu hinterfragen – egal ob sie von einer erfolgreichen CEO oder einem beliebten Influencer kommen.

weiterlesen …

Vielen Dank, dass Du Dir die Zeit genommen hast, diesen Newsletter zu lesen. Ich hoffe, die Inhalte konnten Dich inspirieren und Dir wertvolle Impulse für Dein (digitales) Leben geben. Bleib neugierig und hinterfrage, was Dir begegnet!

---

EpicMind – Weisheiten für das digitale Leben „EpicMind“ (kurz für „Epicurean Mindset“) ist mein Blog und Newsletter, der sich den Themen Lernen, Produktivität, Selbstmanagement und Technologie widmet – alles gewürzt mit einer Prise Philosophie.

---

Disclaimer Teile dieses Texts wurden mit Deepl Write (Korrektorat und Lektorat) überarbeitet. Für die Recherche in den erwähnten Werken/Quellen und in meinen Notizen wurde NotebookLM von Google verwendet. Das Artikel-Bild wurde mit ChatGPT erstellt und anschliessend nachbearbeitet.

Topic #Newsletter

I hosted a lot of friends over today. They came over from 1 all the way till eight or 9 PM. This was the most amount of people I’ve hosted, like around like 12 or 13 at the peak. I’m glad I did it, but my social battery I think is drained. I think there’s also a certain amount of growing pain in finding your community of people and who you feel comfortable with.

I don’t think about her as much anymore. I almost have started to forget her face, but whenever I remember that, I think back to her and I have to stop myself. I also know that I have a lot of photos of her, and honestly I want to go back and look at them. But I don’t do it cause I know it’s smart not to. It’s strange because I don’t even know how much of her I miss, versus if I just miss the holes that she filled. But at the same time I think it’s a little bit of a mixture of both. I really do miss a lot of the connection in the suite moments that we shared, and a lot of the things that we were able to do together. But I think this is a dangerous thing, romanticizing things this soon. I still find that some of the places where it hurt me a lot are still sore. I try not to avoid things that remind me of her, but I don’t really want to see anything about VALORANT or VR chat. It’s weird to have tied myself so closely to someone who gave me so much doubt and anxiety. It’s weird the lack of self-respect/self-love that I had. I think I wanted a relationship so desperately, and I wanted it to work so badly that I kept telling myself that what happened was just a fluke over and over again. But at the same time that doesn’t make it any different, how nice and safe it felt to wake up to her. In the middle of the night waking up, and rolling over and pulling her arm over me, and getting to be hugged and cuddled to bed. Having someone that would lay on me. Feeling her hands pushing on my body in her shitty attempts at massages. It’s hard because she wasn’t a great partner a lot of different ways, but I think she did try. And at the end of the day someone can be good but not good enough. And I guess I just have a higher bar. I think she will have another partner that is maybe a little bit less mentally dominant that can coexist with her a little bit better, and things can be more her speed. And I think she’ll be happy and I think her partner will be very happy with her. She has a very kind heart. Just a bit naïve and with some growing to do. And I kind of feel like I grabbed her along and dragged her at my speed in life, and I think that’s not something she was ready for. We really are at different stages of life. In more ways than one. And I don’t miss having to almost regress myself in several different ways to match a little bit more. And I really like the stability that I have now. But I do mourn the future that I had planned and hoped for. I kept telling myself that people are just young right now, and if you give it a little bit of time people will mature more and grow. And I think that’s true, but at the same time I don’t know what I can even expect.

It’s a weird thing to me, I feel like in a lot of ways I consider myself exceptional, but at the same time I have problems with my self-worth. Sometimes I just wish that I was loved as a child and I wouldn’t have to worry about trying to figure out a concept that I never could believe. Like how could you fucking tell me that someone could just love you no matter what. Like even if you did bad at things, or even if you fucked up or even if you asked for help or a fucking hug sometimes, they would actually give it to you? Like if you told your parent that you were hurting, they would care? What a stupid fucking fantasy that is. And I know that it’s reality for a lot of people, but I just almost want to refuse the fact that it exists. The grapes I cannot taste must be sour. But I fucking just wish I could have been loved, not even for how much nicer it would’ve made the rest of my life, and maybe not making me try to kill myself. But even just for the fact that I could see myself as someone worth loving to myself, and to others. Because I say that I love myself and I think I do, but at the same time when I think of anyone else it’s almost like the only thing I should be loved for is either value, or loyalty from value I’ve already provided. And that fucking hurts to go through life that way.

この部屋にはもう用がないので、出ていくことにした。 自分の顔はすっかり童話めいていて、ドアから出る行為によく似合っていた。 このドアには最初から鍵はなく、いつでもその気になれば出られたのだ。

ドアに手をかけ、後ろを振り返る。 部屋には大きな木の机と、木の椅子がひとつ。 壁には時計がかかっていた。

住んでいるときは何とも思わなかったが、 出ていくとなると、悪くない部屋だったかもしれないと思えるレイアウトだ。

一度ドアノブから手を離し、小さな部屋をゆっくり一周することにした。

ミシミシと床が音を立てる。 いつもより大きく聞こえる床の音。

呼吸も。 息を吸ったり吐いたり。 意図的に、吸ったあともう一度吸う遊びをしてみたりする。 息をするたび、お腹が暖かくなる。

窓から差し込む光は青白い。 まだ午前中だが、こんなことをしていたら、すぐにオレンジ色の光になってしまうから人生は忙しい。

リュックにさしていた水を飲む。 水というより、口の中の味を感じる。 自分は500mlを基本に飲むが、世の中には2Lのペットボトルに口をつける人もいることを思い出す。 丼が大盛より普通盛りの方が美味しいことは、なんとなくみんな気づいていると思う。 水も同じで、2Lより500mlのほうが美味しい。 ただ、コップに注ぐ場合は2Lから出した方が美味しいけれど。

右足より左足に重心がある。 というより、右足で地面を踏ん張れていない感覚がある。 左足を浮かせ、右足だけで立ってみても、それでも重心は左に寄る。 ただの違和感で、医者は何も言ってくれないだろう。 分かりやすいケガや痛みより、違和感を解決するほうがずっと難しい。

首の後ろに、チップのようなものが埋め込まれている気がする。 こんな質素な部屋に、望んで長くいたのだから、本当に誰かにチップを埋め込まれていてもおかしくない。 しかし、本当にチップが埋め込まれているわけではない。 それでも、そこにチップがあるという感覚。 自分の記憶が重要であることを、自分の脳内で視覚化しているのだ。

遠くで救急車の音がする。 誰かが運ばれていて、その人が助かったのかどうかは、分かる術がない。 毎日救急車の音が響くのに、自分の生活には直接影響がない。 救急車で運ばれる人は、普段は何をしているのだろう。

さて、部屋を一周したので外に出る。 床に落ちているホコリに「それでは」と言い、 悲しい顔は家側に、凛々しい顔は外に向け、 目的地はなく、ただまっすぐ歩くことだけを決めて、その家を跡にした。

A lone traveler embarks on a journey through a vibrant valley of blogging platforms, seeking the perfect path to share their voice and stories with the world.

Okay, it has been one year since I joined Micro.blog and Scribbles.page, and just over a year since I joined Write.as. I thought I would review all three services with a clear winner, a hard “can't wait for my subscription to end and won't be renewing,” and a dark horse.

I joined all three within months of each other looking to get rid of my InMotionHosting web host and get away from WordPress. I didn't like the direction that Matt Mullenweg was heading and didn't want to get burned like I did with Elon Musk and Twitter. Twitter was a special place for me as I refused to use any Zuckerberg product, especially since he ruined Instagram.

Now with the history out of the way, here we go.

Write.as — I joined the free tier in October/November 2024 and was initially impressed by its simplicity compared to WordPress. I like paying for services ahead of time, so I bought the five-year plan. That was buyer's regret.

Customization is where Write.as falls apart. Anything beyond typing and publishing requires contorting CSS and JavaScript, and even then there are limits. The rich editor is buggy and loses formatting if you switch between rich and plain text modes.

The platform feels stagnant. Post preview was first requested in October 2018. Over seven years later, it finally got shipped but within the plain text editor only and does not account for any custom CSS. Users are asking for more than a simple text preview. They want to actually see how the post is going to look live. Support has historically been slow, though the owner has recently brought on some help.

The sole owner has been transparent about his shifting priorities. He took a sabbatical from development in 2022 and has written about moving toward other creative pursuits. In recent blog comments, Matt acknowledged taking mental health breaks “at different points over the years” and has even considered succession planning. While his transparency is commendable, paying customers are left wondering when development will resume in earnest.

Photo integration through Snap.as is frustrating. If you want picture galleries, you have to pay extra, but you can't even embed them in Write.as posts. In September 2025, the owner asked users what they'd want galleries to look like, saying “the design is the biggest thing holding us back.” After years as a paid feature, basic functionality is still missing.

The price has increased from $7 to $9 a month, though the proprietor regularly runs promotions and you can pick up 5 years for $180. For comparison, Micro.blog's $5 plan includes blog hosting, custom domain, cross-posting, native apps, and photo sharing. Their $10 Premium plan offers even more. You get dramatically more features and active development for less money.

Pros:

• Excellent Fediverse integration. Posting, editing, and deleting all sync reliably. Any instance can follow Write.as blogs, including Mastodon, Sharkey, and Misskey.
• Sole owner operated, appealing to those who prefer independent services.
• Simple and minimalist if all you need is to type and publish.
• If you get hosting on sale it is a passable deal as long as you know the limitations.

Cons:

• No proper post preview after seven years of requests.
• Rich editor is buggy and limited.
• Customization requires CSS/JavaScript expertise.
• Features disjointed across separate services with extra charges.
• Development has stagnated.
• Price increased to $9/month while offering less than competitors.
• Owner's priorities have shifted away from the platform.

Micro.blog is almost the opposite of Write.as in all ways, and 90 percent of those differences are positive.

I've already reviewed Micro.blog extensively in this blog post here, so I won't rehash everything here. The premium plan is only $1 more than Write.as ($10 vs $9), but you get dramatically more value. Micro.blog is constantly evolving, and the owner maintains development pace while keeping the platform stable and minimally disrupted. Also, keep in mind every tier has different features.

Pros:

• Active weekly development with new features, bug fixes, and improvements shipped constantly
• Extensive feature set: podcasts, newsletters, photo galleries, image hosting, video hosting, multiple blogs, and audio transcription
• Cross-posting to 10+ platforms (Mastodon, Bluesky, Medium, LinkedIn, Tumblr, Flickr, Nostr, Pixelfed, Threads, PeerTube)
• Better customization: themes, custom CSS/JavaScript, Hugo-based with plug-ins
• Import from WordPress, Medium, Ghost, Substack, Write.as, Instagram, Twitter archives, and more
• Responsive support with dedicated Community Manager
• Multiple pricing tiers ($5 basic, $10 premium, $20 Studio) offering more features than Write.as at comparable or lower prices
• Strong community guidelines with human-curated “Discover” section

Cons:

• Awkward Fediverse implementation not fully compatible with non-Mastodon instances
• Features not easily discoverable
• Android apps lagging behind Mac/iOS apps

I would easily recommend this service. It is probably the most well-rounded and actively maintained platform out there if you need these features.

Scribbles.page is my dark horse. This is a managed blog hosting service with excellent design. Vincent Ritter, the owner and designer, has been on a tear lately modernizing the platform and adding features.

The only drawback for me is the lack of Fediverse integration and POSSE. But it makes up for it in every other respect and serves as a nice companion to Micro.blog with built-in cross-posting support.

Vincent is developing a robust API based on JSON and Micropub standards. The only thing I see missing is media uploading, which he is still working on. The pace of changes on Scribbles has been steady and everything is polished.

A social feature unique to the platform is something Vincent calls “Scribbles,” which lets readers send short private messages to blog owners about their posts. It's more casual than email and completely privacy-friendly since scribbles are private notes between the sender and recipient, not publicly shared. The platform also features a nice explorer page where you can discover other blogs, and it's available via RSS feed. Vincent regularly announces software updates there, keeping users informed about new features and improvements. If I had found this before Write.as or Micro.blog, this might have been my only purchase, and the Fediverse could have been implemented via n8n, IFTTT, or a custom solution.

I also appreciate that he plans to offer self-hosting for Lifetime members, and there is a Lifetime membership option instead of subscriptions, which addresses my subscription fatigue. One last detail that might matter to some: it is hosted and based in Europe.

The Verdict

After one year on all three platforms, here's my decision:

Write.as is the “won't be renewing.” Unless it drastically changes course in the next five years, it is too limited and stagnant. While the Fediverse integration is excellent, that alone doesn't justify the price when competitors offer more features and active development. Only consider it if you get a significant promotional discount and need nothing beyond basic blogging with ActivityPub.

Micro.blog is the clear winner. It delivers exceptional value with constant development, extensive features, and strong community management. The platform continues to evolve while remaining stable. Despite everything increasing in price lately, I am surprised Micro.blog hasn't raised its rates. I wholeheartedly recommend it.

Scribbles.page is the dark horse. If you don't need federation features and value gorgeous design with modern blogging standards, this is a compelling choice. The lifetime membership option and Vincent's impressive development momentum make it worth serious consideration.

---

Links may be shortened via mtribe.link for cleaner formatting. All links redirect to their original destinations.

#opinion #review

Somewhere inside a foundation model trained on millions of supposedly de-identified electronic health records, a ghost lingers. Not a literal one, of course, but a data spectre: the clinical history of a patient whose records were stripped of names, addresses, and social security numbers before ever touching an algorithm. The model was never supposed to remember this person. It was supposed to learn medicine. Instead, it learned a patient.

This is the memorisation problem, and it is rapidly becoming one of the most consequential privacy challenges in clinical artificial intelligence. As healthcare systems worldwide rush to deploy foundation models trained on vast troves of electronic health record data, researchers are discovering that de-identification, the process long treated as the gold standard for protecting patient privacy, may not be enough. These models do not merely generalise medical knowledge from the populations they study. In some cases, they memorise individual patient records with enough fidelity that an adversary armed with the right prompts could extract sensitive clinical details about real people.

The implications are profound. A patient with a rare autoimmune disorder, an individual whose HIV status was recorded during a hospital visit, a person who sought treatment for substance use: these are precisely the kinds of patients whose records are most vulnerable to memorisation, because their clinical profiles are, by definition, unusual. And unusualness is exactly what makes data memorable to a machine learning model.

When Models Remember What They Should Forget

In October 2025, a team of researchers led by Sana Tonekaboni, a postdoctoral fellow at the Eric and Wendy Schmidt Center at the Broad Institute of MIT and Harvard, published a paper that would reframe how the clinical AI community thinks about privacy. The study, “An Investigation of Memorization Risk in Healthcare Foundation Models,” was presented at the 2025 Conference on Neural Information Processing Systems (NeurIPS) in San Diego. Co-authored with Lena Stempfle, Adibvafa Fallahpour, Walter Gerych, and Marzyeh Ghassemi, an associate professor at MIT in Electrical Engineering and Computer Science, the paper introduced a suite of black-box evaluation tests designed to probe whether foundation models trained on structured electronic health records were genuinely generalising medical knowledge or simply recalling individual patients.

The distinction matters enormously. A model that generalises has learned, say, that patients over 65 with elevated troponin levels and chest pain are at high risk of myocardial infarction. That knowledge draws on thousands of patient records and reflects a genuine population-level pattern. But a model that memorises has locked onto a singular patient record, and when prompted with the right combination of attributes, it can reproduce details about that specific individual. “Knowledge in these high-capacity models can be a resource for many communities,” Tonekaboni explained, “but adversarial attackers can prompt a model to extract information on training data.”

The framework the team developed includes methods for probing memorisation at both the embedding level, where models encode patient data as numerical representations, and the generative level, where models produce clinical outputs. Crucially, the researchers designed their tests to distinguish between benign generalisation and genuinely harmful memorisation. Not all information leakage is created equal. If a model reveals that a particular patient profile tends to involve elderly males, that reflects population statistics. If it reveals that a specific combination of laboratory values, timestamps, and diagnostic codes corresponds to a single identifiable individual, that constitutes a privacy breach.

The findings were sobering. The researchers demonstrated that the more prior knowledge an attacker possesses about a particular patient, the more likely the model is to leak additional information. Patients with rare conditions proved especially vulnerable, precisely because their clinical signatures are distinctive enough to be picked out from the broader training distribution. And while some categories of leaked information, such as a patient's age or gender, represent relatively low risk, others carry serious consequences. Diagnoses related to HIV, substance use disorders, or mental health conditions were flagged as potentially harmful disclosures that could damage a person's employment prospects, insurance coverage, or social standing.

Ghassemi, the paper's senior author, offered a practical framing of the threat. “We really tried to emphasise practicality here,” she noted. “If an attacker has to know the date and value of a dozen laboratory tests from your record in order to extract information, there is very little risk of harm. If I already have access to that level of protected source data, why would I need to attack a large foundation model for more?” The question cuts to the heart of the adversarial calculus: how much prior knowledge makes an attack feasible, and at what point does memorisation cross from theoretical vulnerability to practical danger?

The Adversary's Toolkit

To understand the scale of the memorisation threat, it helps to look beyond healthcare-specific models to the broader landscape of large language model security research. The foundational work in this space comes from Nicholas Carlini and colleagues, whose research at Google DeepMind and collaborating institutions has systematically demonstrated that language models memorise and can be made to regurgitate their training data.

In a landmark 2021 paper published at USENIX Security, Carlini, along with Florian Tramer, Eric Wallace, and others, showed that an adversary could extract hundreds of verbatim text sequences from GPT-2, including personally identifiable information such as names, phone numbers, and email addresses. The attack required no access to the training data itself, only the ability to query the model. By 2023, the same research group, now including Milad Nasr, Daphne Ippolito, and Christopher Choquette-Choo, had scaled their methods dramatically. Their paper “Scalable Extraction of Training Data from (Production) Language Models” demonstrated that an adversary could extract gigabytes of training data from both open-source models and commercial systems including ChatGPT.

The 2023 work introduced a particularly concerning technique: the divergence attack. By crafting prompts that cause a model to diverge from its normal conversational behaviour, the researchers achieved training data emission rates up to 150 times higher than those observed during typical usage. The attack essentially tricks aligned models into reverting to their pre-alignment behaviour, at which point they begin outputting memorised sequences with alarming fidelity.

What does this mean for clinical AI? The attack surface is substantial. An electronic health record foundation model trained on millions of patient records contains, by design, sensitive clinical information. Even if the records have been de-identified according to HIPAA standards, the model itself may have encoded enough information to reconstruct individual patient profiles when queried with the right combination of clinical attributes. A rare diagnosis combined with a specific age range and a distinctive pattern of laboratory values could function as a fingerprint, allowing an attacker to extract additional details that the de-identification process was supposed to protect.

The level of prior knowledge required for a successful attack varies depending on the model architecture, training methodology, and the patient population in question. Research on general-purpose language models suggests that model size strongly correlates with memorisation: larger models, with their greater capacity to store training data patterns, are more vulnerable to extraction attacks. Given that clinical foundation models are trending towards ever-larger architectures to capture the complexity of medical knowledge, this scaling relationship poses a direct tension between clinical utility and patient privacy.

De-identification Was Never Bulletproof

The memorisation problem does not exist in isolation. It builds upon decades of research demonstrating that de-identification of health data has always been more fragile than regulators and healthcare institutions have assumed.

The seminal work in this field belongs to Latanya Sweeney, now the Daniel Paul Professor of the Practice of Government and Technology at the Harvard Kennedy School. In 1997, while still a graduate student at MIT, Sweeney demonstrated that she could re-identify the medical records of then-Massachusetts Governor William Weld by cross-referencing publicly available voter registration data with de-identified hospital discharge records. The records had been stripped of names, addresses, and social security numbers, but they retained date of birth, gender, and ZIP code. Sweeney showed that just these three attributes were sufficient to uniquely identify an individual.

Her subsequent research revealed that 87 per cent of the United States population could be uniquely identified using only ZIP code, date of birth, and gender, a finding that helped shape the HIPAA Privacy Rule's Safe Harbour de-identification standard. Yet even with these protections in place, re-identification remains possible. A 2018 study demonstrated that patients could be re-identified from HIPAA-compliant de-identified datasets by cross-referencing them with publicly available newspaper articles about hospitalisations.

A 2025 paper published in AI and Ethics highlighted the particular challenge of clinical free text. Structured data fields like diagnosis codes and laboratory values can be systematically scrubbed, but clinical notes contain narrative descriptions that may include identifying details embedded in the prose: references to a patient's occupation, family circumstances, or the name of a referring physician. De-identification tools, including those powered by natural language processing, struggle with the ambiguity and variability of clinical language.

The emergence of foundation models adds a new dimension to this longstanding vulnerability. Traditional re-identification attacks required an adversary to obtain and cross-reference multiple external datasets. Memorisation attacks against AI models require only the ability to query the model itself. The model becomes both the target and the pathway to the data it was trained on, collapsing what was previously a multi-step process into a single interaction. A 2025 study published in PMC on contemporary threats to anonymised healthcare data warned that AI-based techniques can now infer identity from traditionally de-identified sources using data such as electrocardiograms or patterns of gait, data types that were never considered identifiers under existing privacy frameworks.

How AI Threats Compare with Conventional Cybersecurity Risks

The memorisation vulnerability exists within a broader landscape of healthcare cybersecurity threats that are already severe and worsening. Understanding how AI-specific risks compare with conventional attack vectors is essential for calibrating the response.

The numbers from conventional healthcare cybersecurity are staggering. In 2024, 259 million Americans had their protected health information compromised through hacking incidents, a figure driven overwhelmingly by the Change Healthcare ransomware attack. That single breach, perpetrated by the ALPHV/BlackCat ransomware group, affected approximately 190 million individuals after attackers exploited a Citrix remote access service that lacked multi-factor authentication. UnitedHealth Group, Change Healthcare's parent company, reported total cyberattack impacts of 2.457 billion dollars in the first nine months of 2024 alone.

The healthcare sector has become the most targeted industry for ransomware, accounting for 17 per cent of all ransomware attacks across sectors. Complete protected health information packages command prices of up to 1,200 dollars per record on criminal marketplaces, roughly 80 times the value of stolen credit card data. Over 80 per cent of stolen health records in 2024 were taken not from hospitals directly but from third-party vendors, software services, and business associates, highlighting the systemic nature of the vulnerability.

Against this backdrop, AI memorisation attacks represent a qualitatively different kind of threat. Conventional breaches involve exfiltrating stored data, breaking through perimeters, and exploiting network vulnerabilities. Memorisation attacks exploit the model itself as an unwitting data store. There is no firewall to breach, no database to penetrate. The sensitive information is encoded within the model's parameters, distributed across billions of numerical weights in ways that resist simple detection or removal. An attacker needs nothing more than API access to the model, which in many clinical deployment scenarios would be available to any authorised user of the system.

The two categories of threat also differ in their detectability. A ransomware attack produces obvious signs: encrypted systems, operational disruption, ransom demands. A memorisation extraction attack can be conducted through queries that resemble normal clinical usage, making it far harder to detect. Medical identity theft already takes an average of 24 months to discover, compared with four months for financial fraud. Memorisation-based data extraction could extend this detection timeline even further, because the data never technically leaves the system in the conventional sense.

Yet it would be a mistake to treat AI memorisation as the dominant threat. The scale of conventional breaches dwarfs anything that memorisation attacks have demonstrated in practice. The Change Healthcare incident compromised the records of roughly 190 million people in a single event. Memorisation attacks, by contrast, tend to target individual patients or small groups, requiring specific prior knowledge about each target. The threat from memorisation is more surgical than it is sweeping, but for the individuals affected, particularly those with rare conditions or stigmatising diagnoses, the consequences could be devastating.

The Regulatory Patchwork

The regulatory response to AI memorisation risks in healthcare remains fragmented and, in many respects, inadequate. Existing frameworks were designed for a world where privacy threats came from databases, not algorithms.

In the United States, HIPAA remains the foundational framework for protecting health information, but it was enacted in 1996, long before the emergence of clinical AI. The proposed update to the HIPAA Security Rule, published by the Department of Health and Human Services in January 2025, represents the first major revision in over a decade. The proposal eliminates the distinction between “required” and “addressable” security controls, mandates encryption for all electronic protected health information, and introduces multi-factor authentication requirements. Critically, it establishes that data used in AI training, prediction models, and algorithm development by regulated entities falls under HIPAA's protections.

However, the proposed rule does not specifically address memorisation risks. It treats AI systems primarily through the lens of conventional cybersecurity: access controls, encryption, audit logging. These measures are necessary but insufficient for a threat that is embedded within the model's learned representations rather than stored in a conventional database. The public comment period for the proposed rule closed in March 2025 with nearly 5,000 submissions, and the final rule is expected in late 2025 or 2026. Whether it will address the unique characteristics of AI memorisation remains uncertain.

The European Union's approach through the AI Act offers somewhat more specificity. The regulation classifies AI systems used in healthcare as high-risk, subjecting them to requirements for data governance, transparency, human oversight, and post-market monitoring. From August 2026, most obligations will apply, with full compliance for high-risk medical device AI required by August 2027. The Medical Device Coordination Group published guidance document MDCG 2025-6 to clarify how the AI Act interacts with existing medical device regulations under the MDR and IVDR frameworks.

The AI Act's data governance requirements are particularly relevant to memorisation. High-risk AI manufacturers must implement practices appropriate for the intended purpose, including attention to possible biases and privacy risks. The transparency obligations require that systems be designed to allow deployers to interpret outputs and use systems appropriately. These provisions create a regulatory foundation that could, in principle, require memorisation testing before deployment. But the specifics of implementation remain to be worked out through standards and guidance that have not yet been finalised.

At the state level in the United States, a patchwork of legislation is emerging. By 2025, over 250 AI-related bills had been introduced across more than 34 states. Texas enacted the Responsible Artificial Intelligence Governance Act in June 2025, requiring healthcare practitioners to provide patients with written disclosure of AI use in diagnosis or treatment. Colorado and Utah have enacted their own comprehensive AI laws. The result is a fragmented landscape that creates compliance challenges for healthcare organisations operating across jurisdictions whilst providing inconsistent protection for patients.

Technical Safeguards and Their Limits

The technical toolkit for mitigating memorisation risks is growing, though no single approach offers a complete solution.

Differential privacy, the mathematical framework developed by computer scientists including Cynthia Dwork of Harvard University, provides formal guarantees about information leakage during model training. By adding carefully calibrated statistical noise to the training process, differential privacy ensures that the model's outputs reveal almost nothing about any individual training example. Recent research has demonstrated that healthcare AI models can achieve 96.1 per cent accuracy with a privacy budget of epsilon equals 1.9, suggesting that strong privacy and high clinical performance can coexist.

Yet differential privacy has limitations. The privacy-utility trade-off is real: stronger privacy guarantees require more noise, which can degrade model performance on clinical tasks where accuracy directly affects patient outcomes. The United States Census Bureau's experience with differential privacy in the 2020 census provides a cautionary example. Research found that the technique introduced disproportionate discrepancies for rural and non-white populations, raising concerns about equity impacts that would be equally relevant in clinical settings where underrepresented populations already face disparities in care.

Federated learning offers another approach, keeping patient data decentralised across institutions whilst training a shared model. Rather than aggregating raw data on a central server, each participating hospital trains the model locally and shares only model updates. Yet research has shown that these model updates themselves can leak information. Gradient inversion attacks can reconstruct substantial portions of original training data from the mathematical updates exchanged during federated learning. A study titled “Two Models are Better than One: Federated Learning Is Not Private for Google GBoard Next Word Prediction” demonstrated that user sentences could be reconstructed from model updates alone.

Machine unlearning, the targeted removal of specific patient data from a trained model, has emerged as a conceptually appealing response to memorisation. The approach aligns with the General Data Protection Regulation's right to be forgotten, which allows individuals to request deletion of their personal data. Research presented at MICCAI 2025 introduced Forget-MI, a method for unlearning multimodal medical data from trained architectures. A December 2025 testbed called MedForget modelled hospital data as a nested hierarchy, enabling fine-grained unlearning assessment across multiple organisational levels.

But machine unlearning faces fundamental practical barriers. Retraining a model from scratch without specific patient data remains the only guaranteed path to complete unlearning, and for large foundation models, retraining can take weeks and cost millions of dollars. Approximate unlearning methods are faster but cannot guarantee that all traces of a patient's data have been removed. Moreover, if certain demographic groups are more likely to exercise their right to be forgotten, the resulting training data could become skewed, potentially worsening the very biases that clinical AI is supposed to help address. As a Health Affairs analysis noted, machine unlearning “is computationally intensive, scientifically immature, and potentially destabilising to models that must remain reliable across a wide range of clinical inputs.”

Data deduplication, the removal of repeated training examples, provides a simpler but partial mitigation. Research has consistently shown that models are more likely to memorise data that appears multiple times in training sets. Curating and deduplicating training data can reduce memorisation rates, though it cannot eliminate the risk entirely for patients whose clinical profiles are inherently distinctive.

Building an Evaluation Framework That Works

The MIT team's work points towards what a comprehensive evaluation framework for clinical AI memorisation might look like. Their open-source toolkit, validated on a publicly available electronic health record foundation model, provides a starting point for systematic privacy assessment before model deployment.

The framework's key innovation is contextualising memorisation within healthcare. Not all information leakage constitutes a meaningful privacy risk. A model that reveals population-level patterns, such as the typical age distribution of patients with a particular condition, is doing exactly what it was designed to do. The danger arises when a model's outputs can be traced to a specific individual, particularly when the leaked information includes sensitive diagnoses or treatment histories.

Tonekaboni emphasised the importance of practical evaluation. “This work is a step towards ensuring there are practical evaluation steps our community can take before releasing models,” she said. The framework assesses both embedded memorisation, where patient information is encoded in the model's internal representations, and generative memorisation, where the model can be prompted to produce patient-specific outputs. By testing across both dimensions, the framework provides a more complete picture of privacy risk than either approach alone.

For this kind of evaluation to become standard practice, it would need to be integrated into the regulatory approval process for clinical AI systems. Currently, most AI-enabled medical devices in the United States are cleared through the FDA's 510(k) pathway, which requires demonstration of substantial equivalence to a previously approved device but does not mandate independent clinical performance studies or privacy evaluation. A cross-sectional study of 903 FDA-approved AI devices found that clinical performance studies were reported for only approximately half at the time of regulatory approval. Memorisation testing is not part of the approval process at all.

The Coalition for Health AI (CHAI), on whose working group Ghassemi serves, represents one effort to establish industry-wide standards for trustworthy health AI. The NIST AI Risk Management Framework provides a complementary structure, addressing validity, reliability, safety, security, explainability, privacy, and fairness. Integrating memorisation evaluation into these existing frameworks would be more practical than creating entirely new regulatory apparatus, but it requires agreement on what constitutes acceptable levels of memorisation risk, a question that remains open.

Rare Conditions, Outsized Vulnerability

The memorisation problem falls hardest on the patients who can least afford it. Individuals with rare diseases, by definition, have clinical profiles that stand out from the broader population. Their diagnostic codes appear infrequently in training data. Their laboratory value patterns are unusual. Their treatment trajectories are distinctive. All of these characteristics make their records more memorable to a model and more extractable by an adversary.

The same is true for patients with stigmatising diagnoses. HIV status, substance use disorders, psychiatric conditions, and sexually transmitted infections all carry social consequences that extend far beyond the clinical encounter. Disclosure of these conditions can affect employment, housing, insurance, and personal relationships. De-identification was supposed to sever the link between these sensitive details and the individuals they describe. Memorisation threatens to re-forge that link through the model itself.

This disproportionate vulnerability raises equity concerns that mirror broader patterns in healthcare AI. Research has repeatedly shown that AI systems can perpetuate and amplify existing biases against marginalised populations. If memorisation risks are concentrated among patients with rare or stigmatising conditions, the privacy burden falls most heavily on those who are already underserved by the healthcare system.

Addressing this inequity requires targeted protections. Higher levels of differential privacy noise could be applied to training data involving sensitive diagnoses, at the cost of reduced model performance for those specific conditions. Rare disease patient records could be excluded from training sets entirely, though this would eliminate the clinical utility of foundation models for precisely the populations that stand to benefit most from AI-assisted care. Neither option is satisfactory, and the tension between privacy protection and clinical benefit for rare disease patients may prove to be one of the defining challenges of clinical AI governance.

What Genuine Protection Requires

The path from current vulnerability to genuine protection requires action across multiple domains simultaneously. No single technical safeguard, regulatory standard, or evaluation framework will suffice in isolation.

On the technical side, differential privacy during training should become the default rather than the exception for clinical foundation models. Memorisation evaluation, using frameworks like the one developed by Tonekaboni and colleagues, should be mandatory before any model is deployed in a clinical setting. Ongoing monitoring should be built into deployment infrastructure to detect potential memorisation-based extraction attempts in real time. And machine unlearning capabilities, however immature, should be developed and standardised so that patients can exercise meaningful control over the fate of their data within AI systems.

On the regulatory side, HIPAA needs to evolve beyond its current framework to address threats that are embedded within model architectures rather than stored in conventional databases. The EU AI Act's high-risk classification for healthcare AI provides a useful starting point, but implementation guidance must specifically address memorisation risks. Regulatory bodies including the FDA, the European Medicines Agency, and national health authorities need to incorporate memorisation testing into their approval and post-market surveillance processes.

On the institutional side, healthcare organisations deploying clinical AI must treat memorisation as a distinct category of risk requiring its own governance structures, audit procedures, and incident response plans. The conventional cybersecurity toolkit, with its emphasis on perimeter defence, encryption, and access control, is necessary but not sufficient for threats that live inside the model rather than outside the firewall.

The researchers behind the MIT study plan to expand their work to become more interdisciplinary, bringing in clinicians, privacy experts, and legal scholars. That instinct is exactly right. The memorisation problem sits at the intersection of computer science, medicine, law, and ethics, and solving it will require all four disciplines working in concert.

“There's a reason our health data is private,” Tonekaboni observed. “There's no reason for others to know about it.” That principle has guided health privacy law for decades. The question now is whether it can survive the age of foundation models trained on the very data it was designed to protect. The answer will depend on whether the clinical AI community treats memorisation as a fundamental design constraint rather than an afterthought, building privacy into the architecture of these systems from the ground up rather than bolting it on after deployment. The technology to do so exists. Whether the will and the regulatory momentum exist to mandate it remains the open question.

---

References and Sources

1. Tonekaboni, S., Stempfle, L., Fallahpour, A., Gerych, W., and Ghassemi, M. “An Investigation of Memorization Risk in Healthcare Foundation Models.” arXiv:2510.12950, presented at NeurIPS 2025. https://arxiv.org/abs/2510.12950

2. MIT News. “MIT scientists investigate memorization risk in the age of clinical AI.” January 5, 2026. https://news.mit.edu/2026/mit-scientists-investigate-memorization-risk-clinical-ai-0105

3. Carlini, N., Tramer, F., Wallace, E., et al. “Extracting Training Data from Large Language Models.” USENIX Security 2021. https://www.usenix.org/conference/usenixsecurity21/presentation/carlini-extracting

4. Nasr, M., Carlini, N., Hayase, J., et al. “Scalable Extraction of Training Data from (Production) Language Models.” arXiv:2311.17035, 2023. https://arxiv.org/abs/2311.17035

5. Sweeney, L. “Simple Demographics Often Identify People Uniquely.” Carnegie Mellon University, Data Privacy Working Paper 3, 2000. https://dataprivacylab.org/people/sweeney/work/index.html

6. Sweeney, L. “Risks to Patient Privacy: A Re-identification of Patients in Maine and Vermont Statewide Hospital Data.” Technology Science, 2018. https://techscience.org/a/2018100901/

7. PMC. “Addressing contemporary threats in anonymised healthcare data using privacy engineering.” 2025. https://pmc.ncbi.nlm.nih.gov/articles/PMC11885643/

8. Springer Nature. “What is the patient re-identification risk from using de-identified clinical free text data for health research?” AI and Ethics, 2025. https://link.springer.com/article/10.1007/s43681-025-00681-0

9. HIPAA Journal. “Healthcare Data Breach Statistics.” https://www.hipaajournal.com/healthcare-data-breach-statistics/

10. UnitedHealth Group. “UnitedHealth Group Updates on Change Healthcare Cyberattack.” April 22, 2024. https://www.unitedhealthgroup.com/newsroom/2024/2024-04-22-uhg-updates-on-change-healthcare-cyberattack.html

11. HHS. “Changes Proposed to Strengthen HIPAA Security Rule.” January 2025. https://www.hhs.gov/hipaa/for-professionals/special-topics/de-identification/index.html

12. Reed Smith. “The EU AI Act and Medical Devices: Navigating High-Risk Compliance.” 2025. https://www.reedsmith.com/our-insights/blogs/viewpoints/102kq35/the-eu-ai-act-and-medical-devices-navigating-high-risk-compliance/

13. European Commission. “Medical Devices Joint Artificial Intelligence Board, MDCG 2025-6.” 2025. https://health.ec.europa.eu/document/download/b78a17d7-e3cd-4943-851d-e02a2f22bbb4_en

14. Health Affairs Forefront. “Unlearning In Medical AI: A New Frontier For Privacy, Regulation, And Trust.” 2025. https://www.healthaffairs.org/content/forefront/unlearning-medical-ai-new-frontier-privacy-regulation-and-trust

15. MICCAI 2025. “Forget-MI: Machine Unlearning for Forgetting Multimodal Information in Healthcare Settings.” https://arxiv.org/html/2506.23145

16. MedForget. “MedForget: Hierarchy-Aware Multimodal Unlearning Testbed for Medical AI.” December 2025. https://arxiv.org/html/2512.09867v1

17. Nature Medicine. “Medical large language models are vulnerable to data-poisoning attacks.” January 2025. https://www.nature.com/articles/s41591-024-03445-1

18. Becker's Hospital Review. “EHR-trained AI could compromise patient privacy: MIT.” 2026. https://www.beckershospitalreview.com/healthcare-information-technology/ai/ehr-trained-ai-could-compromise-patient-privacy-mit/

19. Cobalt. “Healthcare Data Breach 2025 Statistics.” https://www.cobalt.io/blog/healthcare-data-breach-statistics

20. NIST AI Risk Management Framework. https://www.nist.gov/artificial-intelligence/ai-risk-management-framework

---

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk