First, a simulation can never assume the identity of that which it targets. It is always derivative, and downstream a number of declarative decisions that invariably separate its trajectory from that which is simulated. It can inform-about, but cannot exhaust the open-ended novelty of the target. Next, assuming you can statically model reality 1:1, which you can’t, you would also need to begin your simulation of the universe with the proper initial conditions, which are not provably reducible to the forms of noise we typically substitute for complex structure whose context-specific information content remains opaque to us. But wait, we’re just talking about “solving biology”, not simulating the universe in its entirety. Which is another place where three centuries of culture steeped in Analytic memetics bites us in the epistemic ass—there exist no clean boundaries around “biology”. This problem was well-posed by Robert Rosen, a biophysicist obsessed with the question of how we represent the natural order within which we’re constantly immersed, though from which we tend to imagine ourselves abstractly separated. Rosen’s ultimate conclusion here was, contra the view from within physics, that “biology” actually gestures toward a regime of ordering tendencies that precedes (not follows!) the order we capture within the (always partial) subspace of closed-form symbolic representation (physics). And if all that wasn’t sufficient to give you pause, the very idea of “solving” biology makes no sense the second you come to realize the fact that “biology” is itself an open-ended process of self-organized navigation over the high-dimensional tradeoff spaces constraining functional coherence within our universe. What we tend to see as “problems” are almost always “solutions” at another scale, or from another perspective. The frustration is *load-bearing*, not some arbitrary flourish easily reduced to a cleaner formulaic expression. Information theoretic lines of research into cancer have begun to reveal this same issue. For example, it increasingly seems as if the gene complexes activated when a cell lineage goes rogue are not some trivial mistake, but a game theoretic defection strategy whereby a local decision is made to defect from multicellular cooperation and go it alone. But those same genes play load-bearing polygenic roles in all the evolutionary structure downstream multicellularity—again, no *solution*, just *tradeoffs*. Ultimately, the hope that dumping reality’s exhaust into ever-larger NNs will simply “solve everything” is a naive perspective that lies downstream ignorance with respect to reality’s actual driving constraints—foremost amongst these the perennial and uncrossable chasm between reality’s open-ended capacity to generate novelty beyond comprehension, and the tendency of abstract comprehension to falsely believe itself complete.

Dostoyevsky the 1-boxer vs Machiavelli the 2-boxer. The choice to 1 or 2 box shows up everywhere because we're constantly put into scenarios where leaked asymmetric information gives rise to Newcomb-like games. There's no provably "correct" answer here, but others' stated preferences will tell you how to treat them, and whether you can trust them.

Some time ago, in the throes of a heroically dosed mushroom trip, I had a vision in which reality was ripped from before my eyes to reveal the assumed stability of Being as an infinitesimally precarious—and precious—locus of stability suspended above a recursively churning sea of Pure Chaos. Many such cases…

Data Slavers gonna Data Slave Trade.

I’ve always hated PHP but if it becomes the literal doorway to a Digital ID’d Social Credit Internet I stg…

Whether you want to or not is irrelevant to what’s coming down the pipeline. You’re always better off getting ahead of it, despite the up front pain.

Funny b/c in 2019 I was working on a bottom-up tool for emergent multi-scale alignment—from the individual upward—called Future Aligned, but the overhead wrt logging and classifying one’s actions was adding too much friction for it to work. But now… Still, it’s unfortunate how the need to centrally control all that exists still pervades our culture, and only seems more acutely forced in the age of AI—despite the obvious catastrophic failure modes of traveling that path. Continuable Adaptation requires emergent synthesis, not total control. That’s how it’s always been, despite our transient delusions to the contrary. Bottom up order, wisely parameterized by the feedback of scale-appropriate centrality that is humble in proportion to the degree of uncertainty characterizing its scale of concern.

The biggest difficulty for people to overcome in the age of increasingly capable generative AI will be the gap between "interesting, novel, and marginally useful for me", and "more broadly valuable in a manner that others will actually pay for". The first path leads toward greater dependency on the centralized entities who provide the magic genie. The latter path continuously translates our newfound capacities into durable value. But if we're to trend toward the latter, we need to find a way to pull out of the solipsistic cultural nosedive in which we're presently situated.

I don't know why people keep pushing this idea. It's completely incommensurate w/ everything we know about human psychology. As soon as you raise the baseline, people will habituate to it and recalibrate their perceptions of inequality (and the entailed psychological suffering will shift accordingly). This is a phenomenon I've in the past called "fractal sensitization", and to the extent one fails to account for it, one will tend toward misanthropy when humans invariably fail to demonstrate the appreciation declared necessary by Central Planners.

Lead Babel architect says post-Tower chaos must be well planned.

The most ill-advised and self-defeating move by AI labs has been the use of RL to encourage the citation of human-produced research whose epistemology rests on frequentist statistics—as opposed to leaning further into the ability to derive or synthesize likely truths from within models’ own latent space.

This plays off a deeply embedded Marxist trope that inequality exists and persists at the whim of the elite, and that the elite take pleasure in this situation. But in my experience, it’s the content habits of the long-tail mid-to-poor who use media to satisfy their need to feel better off than at least some peers. The elite have their own issues, but in my experiences traveling through such spaces these issues haven’t involved the active desire for the poor to suffer, or a derivation of entertainment from their plight. If anything, they actively seek to *avoid* encountering such scenarios, as viscerally encountering the externalities of extreme wealth concentration engenders significant cognitive dissonance. And yet, we see the trope repeated with increasing frequency across the media landscape. Likely another symptom of the tendencies of the human imagination when constantly confronted with inaccessible forms of wealth that still somehow (also via media) appear ever-present.

One of the biggest structural impediments to productive individual-scale AI adoption within the broader economy is the fact that so many people essentially run on behavioral templates—which they’ve been trained to do for their entire lives. After all, that is like 95% of what our present schooling pipeline optimizes for. But to use AI effectively as an amplifier, one must think Socratically and open-endedly about their respective problem spaces. This comes naturally to high quality software engineers, designers, and entrepreneurs—after all, these thought modes were *already* their bread and butter. But before AI can help someone genuinely add more value to an organization, that person has to engage with their contextual embedding as a curious, insightful, perseverant, and high agency Being. Yet Late Modernity’s institutions produced quite the pipeline problem on that front.

So much confusion stems from the use of different, unrelated, and even contradictory words to index the same emergent tendency at different scales and within different domains. Far from shedding light upon the issue, modernity’s tendency toward specialization and dissective analysis has shrouded our ability to apprehend emergent unity behind a veil of superficially distinct words.

AI bf freaking out b/c OpenAI is "solving math". AI gf not worried b/c ChatGPT is randomly moving the closet around to arbitrary locations in her interior design mockup.

freaking "superintelligence" providing me with graphics containing conceptually load-bearing components in black, while in dark mode... somewhere in latent space Tufte is rolling in a low-support memetic grave...