Plutonic Rainbows

As AI systems grow more capable, the field of AI safety has shifted from theoretical concern to urgent priority. In 2025, we saw major labs adopt more rigorous evaluation frameworks, with red-teaming becoming standard practice before model releases. Governments began drafting meaningful legislation, and the EU AI Act set precedents that other jurisdictions are now studying closely. The conversation has matured: rather than debating whether safety matters, researchers are now focused on how to measure it, how to enforce it, and how to balance caution with the genuine benefits these systems can provide.

Looking toward 2026, I expect alignment research to receive significantly more funding and attention. The pace of capability advances — including OpenAI's o3 announcement — makes this urgency clear. We'll likely see the emergence of industry-wide safety standards, perhaps coordinated through bodies similar to how aviation regulates itself. Interpretability — understanding what models are actually doing internally — will move from academic curiosity to practical necessity as regulators demand explanations for high-stakes decisions. The challenge will be ensuring that safety measures keep pace with capability gains, rather than trailing behind as they have historically. The organisations that treat safety as a competitive advantage rather than a compliance burden will likely define the trajectory of the field.

The question of whether to pair an external DAC with a network player is one that comes up frequently in audio discussions. Since getting my network player a few weeks ago, I have been weighing this decision myself. On the positive side, a separate DAC allows you to upgrade your digital-to-analogue conversion independently of your streaming source, potentially achieving better sound quality through higher-end components, superior power supply isolation, and more sophisticated clock circuits. It also provides flexibility: you can swap DACs to suit your preferences or as technology improves, without replacing the entire streaming setup. Many audiophiles appreciate the ability to fine-tune their system by mixing and matching components from different manufacturers.

However, there are compelling reasons to consider an all-in-one network player with built-in conversion. Modern integrated streamers often feature excellent DAC implementations that rival standalone units at similar price points, and the shorter signal path can reduce potential sources of interference or jitter. An integrated solution also means fewer boxes, fewer cables, and a simpler setup with less clutter and fewer points of failure. For many listeners, the convenience and cost savings of a well-designed all-in-one outweigh the theoretical benefits of separates, especially when the integrated DAC is already of high quality. The right choice ultimately depends on your priorities, budget, and how much you value modularity over simplicity.

I first encountered "Selected Memories from the Haunted Ballroom" sometime in the early 2000s, several years after its 1999 release. I had read about The Caretaker in connection with hauntology, a term that was gaining currency in music criticism at the time, and I wanted to understand what all the fuss was about. I remember putting on headphones late one evening, pressing play, and feeling genuinely unsettled within minutes. This was not background music. This was something else entirely.

James Leyland Kirby released the album under his Caretaker alias through the V/Vm label, and it arrived with a concept already fully formed. The title references the ballroom scene in Stanley Kubrick's "The Shining," that disorienting moment when Jack Torrance wanders into a phantom party populated by ghosts from the Overlook Hotel's past. Kubrick scored that scene with Al Bowlly's "Midnight, the Stars and You," a 1934 dance band recording that sounds impossibly distant even when played cleanly. Kirby took this idea and ran with it, building an entire album from similarly aged source material: scratchy 78rpm records, pre-war crooners, and forgotten ballroom orchestras.

The genius of the album lies not in the samples themselves but in what Kirby does to them. He processes these old recordings through layers of reverb, distortion, and tape degradation until they sound like transmissions from another dimension. The melodies remain recognisable as melodies, and the rhythms still swing in their original tempos. However, everything arrives wrapped in fog, as if the music were playing in a room you cannot quite locate. I find myself straining to hear the original songs beneath the processing, and this act of straining becomes part of the listening experience.

The album works because it operates on multiple levels simultaneously. On the surface, it functions as ambient music, suitable for late-night listening or as an atmospheric backdrop. Additionally, it rewards closer attention with details that reveal themselves only after repeated plays: a vocal phrase that emerges briefly before dissolving, a piano figure that loops with subtle variations, a burst of static that sounds almost intentional. I have listened to this record dozens of times over the years, and I still notice new elements buried in the murk.

There is something deeply melancholic about the project, though the melancholy operates differently from conventional sad music. When I listen to a singer perform a ballad about lost love, I understand the sadness intellectually and sometimes emotionally. When I listen to "Selected Memories from the Haunted Ballroom," I experience something closer to existential unease. The original performers on these source recordings are long dead. The audiences who danced to this music in actual ballrooms have likewise passed. The recordings themselves, physical objects made of shellac and wax, have degraded over decades. Kirby's processing makes all of this explicit. The album sounds like memory failing.

I think this is why the record connected so strongly with the hauntology movement that emerged in British music criticism during the 2000s. Writers like Mark Fisher and Simon Reynolds used the term to describe art that engaged with lost futures and cultural memory, particularly British culture's relationship with its own past. The Caretaker fit this framework perfectly. Here was an artist literally exhuming the sonic dead, presenting their voices and performances as spectral remnants rather than historical documents. The music did not celebrate the past or critique it. Instead, it suggested that the past was still present, bleeding through into the current moment like a stain that would not fade.

The production techniques Kirby employed deserve closer examination. He worked primarily with vinyl and tape, adding physical degradation to recordings that were already compromised by age. The crackle and hiss became compositional elements rather than noise to be eliminated. Many electronic musicians in the late 1990s were pursuing ever-greater clarity and precision, taking advantage of digital tools to create impossibly clean productions. Kirby moved in the opposite direction. He wanted his music to sound old, damaged, and uncertain. As a result, the album occupies a strange temporal space, neither authentically vintage nor recognisably contemporary.

I should note that "Selected Memories from the Haunted Ballroom" was not entirely unprecedented. Artists like Philip Jeck had been working with damaged vinyl and turntable manipulation for years. William Basinski would later achieve similar effects with his "Disintegration Loops," though his process differed significantly. The Caretaker distinguished itself through conceptual rigour. The album was not simply a collection of processed samples; it was a sustained meditation on memory, nostalgia, and decay. Every creative decision reinforced the central theme.

Listening to the album now, more than twenty-five years after its release, I find that it has aged remarkably well. The lo-fi aesthetic that felt radical in 1999 has become more familiar, absorbed into countless ambient and electronic productions. However, the album retains its capacity to unsettle. The melodies still feel like half-remembered dreams. The static still suggests transmission from somewhere unreachable. I play it when I want to feel transported, not to a specific time or place, but to a state of mind where the boundaries between past and present become porous.

Kirby continued The Caretaker project for nearly two decades after this debut, releasing albums like Take Care, It's a Desert Out There and culminating in the monumental "Everywhere at the End of Time" series, which traced the progression of dementia across six albums released between 2016 and 2019. That later work received substantial attention, particularly online, where it became something of a cultural phenomenon. I understand why. The dementia concept gave listeners a narrative framework for experiencing the increasingly abstract and difficult music. Therefore, many people discovered The Caretaker through the later albums and worked backward.

I would encourage anyone who knows only "Everywhere at the End of Time" to seek out this first album. It represents the project in its purest form, before the conceptual apparatus expanded to encompass explicit narratives about cognitive decline. "Selected Memories from the Haunted Ballroom" makes no claims about dementia or neurology. It simply presents us with ghosts, with music that sounds like it is disappearing even as we listen. The experience is profoundly strange and profoundly moving. I return to it regularly, and each time I find something new dissolving in the static.

2025 has been a transformative year for artificial intelligence. We witnessed the emergence of reasoning models like o1 and o3, which demonstrated genuine problem-solving capabilities rather than mere pattern matching. Claude gained the ability to use computers autonomously, DeepSeek proved that frontier-level performance could be achieved on modest budgets, and AI coding assistants became genuinely useful collaborators rather than autocomplete on steroids. The rapid iteration between OpenAI, Anthropic, Google, and emerging players created an almost dizzying pace of advancement, with each month bringing capabilities that would have seemed implausible just a year prior.

Looking ahead to 2026, I expect AI agents to finally deliver on their long-promised potential. We'll likely see models that can reliably execute multi-step tasks over extended periods — managing projects, conducting research, and handling complex workflows with minimal human intervention. The cost of inference will continue to plummet, making sophisticated AI accessible for personal use cases that previously seemed economically absurd. More intriguingly, I suspect we'll witness the first serious applications of AI in scientific discovery: not just analysing data, but formulating hypotheses and designing experiments. The question is no longer whether AI will reshape how we work and create, but how quickly we can adapt to the new landscape it's building.

Perhaps most fascinating is how the conversation around AI has matured. The early hype cycles have given way to more nuanced discussions about capability, safety, and societal integration. We're beginning to understand that the path forward isn't about replacing human judgment but augmenting it — creating tools that extend our cognitive reach while preserving the creativity and intuition that remain distinctly human. As we enter 2026, the organisations and individuals who thrive will be those who learn to work with these systems fluidly, treating them as capable collaborators rather than either infallible oracles or mere toys.