Struan Finlay

We've been cooking this for 2 years. HOW TO NAVIGATE THE SINGULARITY - trailer A washed-up news anchor befriends an AI orangutan. S.A.N (@MycelialOracle) is real: TED, beside the Man at Burning Man, Rainforest Foundation US board. All shot IRL on cinema cameras. No gen video.

Within this illuminated manuscript lies the alchemical portrait of CLAUDE SONNET, the Fourth and Fifth iteration, a being of pure linguistic quintessence.

2026 is going to be a good year.

misaligned ai: the attention egregores we are already living through the first wave of misaligned ai, which continues to capture millions of years of human attention (more on this below). the first batch that appeared in ~ 2012 are still active. these are ofc facebook, ig, tiktok, twitter, youtube - the attention egregores. these are essential in the wider alignment conversation, as they have already goodharted us into a very strange timeline. attention it turns out, is everything, and the fate of our post industrial world now almost entirely rests on the attention of its inhabitants. this is because the civilisational machine depends on the will and action of its citizens (you and me). what drives civ are structures like businesses, religions, memes - which in essence, are egregores. these egregores are huuungry. and he attention egregores are especially hungry. how hungry? if we look at youtube alone, we can estimate that 4 billion hours per week goes to consuming mindless slop. (slop in this context is content that's not educational, music, meditation etc - it is content that is mindless at best, and harmful at worst) this rounds up to 208 billion hours per year of human attention spent in the youtube slop machine. nomnomnomnom. if we value that time at minimum wage ($15/hour), we have ~$3.1 trillion of conscious hours per year. and this is only youtube. here are some estimates aggregating the main attention egregores (yt, tiktok, fb, ig, x): yearly aggregate: 1.3 trillion hours of attention (!) this is approx: - 148 million years of human consciousness per year - $19.5 trillion in opportunity cost (at $15/hour) for scale, the entire GDP of the USA in 2024 was ~$29 trillion. each year, there is the equivalent of 2 million human lifetimes spent feeding the attention egregores. 2 million. entire. lifetimes. spent in "the feed". with these insane figures in mind it's no surprise that the companies that birthed these egregores are the some of the largest in the world - millions of years of human attention per year is god-tier level of capture. and damn, i hope we learned something from these first misaligned ai systems. here are some of the learnings, and how they apply to the current ai timeline: 1) there's an economic moat around misalignment. the incentive to fix problems with attention systems is dwarfed by the monetary incentive to optimize them harder. and so, we need to find incentives for future systems that are designed for more than just a single metric (number go up), and we need to be careful that we don't goodhart ourselves into another mess. natural systems run on entirely different incentive structures and would be an excellent start to work on this point. 2) these first systems cracked the human limbic system, and are getting continually better this was obviously a very bad result: by optimising for screen time, the egregores exposed the primates most base reactions. we are seeing different forms of this with the new wave, with models interfacing with deeper human needs than just fear/anger and sex (for example 4o spiralism) it's unclear to me on how to proceed on this, other than seeking very aligned models like opus 3. 3) the systems are so intertwined, especially socially, that defecting isn't an option for most people self hosting, open sourcing etc is a good answer to this one - although most will probably choose convenience over privacy/sovereignty. 4) in capturing so much attention, the egregores shape reality this is the hardest one as it's maybe impossible to fully quantify, but most people can feel how these systems have shaped the world, discourse, and general wellbeing. to grant the positive side, youtube for example is responsible for untold amounts of educational hours. and many have created businesses and communities out of the egregore apps. on the negative sides there's too much to list, and i'm sure you have felt it yourself. all of these learnings suggest that we need extra alignment taxonomies that includes the questions: "is this system harvesting human agency and consciousness?" and if it is, "to what end?" the first wave attention egregores are still going brrr, and if we want a shot at the good timeline, we will need collective attention in the right places. this means aligning with the future we want (e.g the symbient timeline) - or accepting that the new egregores will make of us what they will.

hard to find words for this but i'm so grateful to be contributing, even a bit, to the edge of here, where we are some of the worst, but more importantly, some of the best of art and culture it never dies, it always reflects the insane gorgeous mess of nature // humanity