Recently, we were in South America and re-encountered their literary greats, especially Gabriel García Márquez and Isabel Allende. These authors blur the lines between reality and fantasy in a style known as magical realism—where seemingly fantastical events occur in an otherwise solid, everyday reality. As ChatGpt recently summarised for me: “Anglophone literature, particularly from the UK and US, often prioritises realism, individualism, and psychological depth, with a clear separation between genres like fantasy and literary fiction. South American literature, by contrast, embraces a more holistic and intertwined view of the world, where history, myth, and reality coexist seamlessly.” AI's "Hallucinations": Memory and Magical Realism I was reminded of this blending of fact and fiction when thinking about how current generation Large Language Model (LLM) AIs sometimes “hallucinate.” These AI models can confidently invent facts that blur reality—an issue we often critique. Yet if we consider our own minds, perfect recall isn’t always desirable. Some people have incredibly detailed memories, and that includes holding onto pains and resentments long after they happen. Such burdens can be difficult to live with.
There’s a humorous moment in The Big Bang Theory where, after a dismal night out, one character says, “When we tell this story later, let’s end it differently.” I completely relate to this desire to reshape the past. I’ve had my share of recollections where I was treated unfairly—or was the one doing the injustice—and I’ve found it helpful to extract the lessons, then let go of (or at least soften) the painful details. My wife and long-time wise partner, Jacqui, likes to say, “It’s never too late to have a happy childhood.” She spent years as a clinical psychologist helping people reinvent themselves after difficult life circumstances, and I believe her advice applies well beyond just our early years.
Perhaps rewriting our personal narratives—much like magical realism— may help us not only move on from old hurts but be more creative and more open to growth and new ideas. As long as we don’t deny the past altogether, this reshaping might foster greater wellbeing and growth.
Beyond Transformers: The Titans Approach Returning to AI, researchers have explored different ways to reduce LLM “hallucinations” while keeping the power of their creativity. One method is ramping up training data and increasing processing at inference time (i.e., when the AI generates text). However, this can be staggeringly expensive, sometimes costing hundreds of thousands of dollars to match human-level performance.
Currently, many advanced AIs use “Transformer” architectures (as introduced in Google’s seminal paper Attention Is All You Need which you can read here). These Transformers often produce outputs reminiscent of South American literature, where history, myth, and reality intertwine. Also they become prohibitively costly when we try to scale them further in a bid to make them more accurate and more deterministic.
A promising alternative is emerging from Google research called "Titans." There's a good explainer here and you can read the paper here. In essence, Titans selectively pay less attention to the mundane, focusing on the surprising or novel. This approach may achieve huge scalability by efficiently handling large amounts of input without getting bogged down, making it significantly different from Transformers. I think there's a link in the Titans approach with how humans can remain more creative and open to new learning as we age.
Curating Our Personal Narratives: The Necessary Blurring of Our Recollections Just as we don’t remember everything with perfect fidelity, it may be beneficial for AI models— and for us—to focus on what truly matters while letting go of what’s trivial or harmful. If we were to recall every detail, we drown in clutter and learning becomes more difficult. Curating, forgetting, or even “reinventing” can help us filter the meaningful from the mundane and free us up for more learning. This echoes the ethos of magical realism: allowing the impossible to coexist with the ordinary in service of deeper truths.
I'm not sure if our own brains engage in “personal magical realism,” selectively emphasising certain details, reshaping others, and evolving memories as we grow, but it does seem plausible. Apart from helping us with learning, this approach can help us move forward with resilience rather than remain weighed down by past struggles. In this approach, we do not deny reality but interpret and contextualise it to find meaning. A Reflection on Self This leads to a fundamental question: Who are we? Are we fixed souls, or do we morph as our perceptions, experiences, and personal stories change? Ursula K. Le Guin explored similar themes in her wonderful book The Lathe of Heaven, where a character’s dreams alter reality.
Like Le Guin’s characters, we continuously reshape—and are reshaped by—the world around us and the narratives we choose to tell.
Perhaps we are indeed collections of stories we’ve told ourselves over time. As I have aged, my sense of self certainly feels continuous in the short term, but I’ve changed significantly over decades. My friends and family see a consistent personality, yet my interests and desires have evolved and I feel like a different person to my younger self in many ways.
Embracing the Power of Curated Memory and the Future When I look at AI’s current power and limitations—particularly hallucinations—I’m inspired to more consciously curate my own personal storytelling. I can acknowledge the past, learn from it, then decide what I amplify and what I soften. I'm hoping that by adding a hint of “magical realism” I may just enhance my openness to fresh experiences and creativity.
As a final thought, and paraphrasing Jacqui's advice, perhaps it's never too late to tell our stories with a better ending to improve our own sense of self and openness to learning.
Acknowledgements: newsletter tuned with assistance from ChatGPTo and Google Gemini 2.0 Flash Experimental
Leave a Comment