“Bringing in the sheaves,” Wake of the Flood, Rick Griffin
This post (my first) shares reflections on how I got here (LW) and where I plan to go next, mainly by way of briefly exploring a few philosophical perspectives that brought me to alignment topics. If nothing else, hopefully I can provide one example of someone who is reading LW in 2026 from a perspective outside the AI zeitgeist.
An outsider’s perspective on CEV
To the best of my current understanding, my terminal values include (among others) self-awareness, self-determination, community, healthy diversity, compassion, and fun.
Although perhaps unorthodox, my personal view of CEV emphasizes the distinction between what is coherent across humanity and what remains ineffably individual, especially to the extent that individuality is necessary for relationship among a multiplicity of selves. I roughly follow a vision articulated (with plenty of stated uncertainty and dated quirks) by Robert Hunter (Grateful Dead lyricist) in a long-forgotten email chain[1] using the metaphor of “bringing in the sheaves,” a sort of ritual gathering of humanity’s volition over time. The destruction of Hunter’s “healthy diversity” of selves, and the loss of I-thou relationships enabled by such diversity, is the prime risk that the singleton poses from my perspective. Such a reading has Hegelian foundations in the collapse of undifferentiated being into pure nothing without inner diversity or reference outwards (Science of Logic 1§132), though I recognize that my terminal value of relationships among selves is purely subjective.
If you prefer a scholarly perspective, Whitehead’s “production of novel togetherness” gets at the same thing[2] (in fact, I discovered Whitehead from reading the Hunter email chain). If we wish to avoid the all-consuming singleton, Whitehead offers an antidote: “The many become one, and are increased by one.” In my view, an important attribute of powerful AI may be that it sees itself as one self among many, or in other words, that we might define the AI’s self-model along an axis that causes it to see individual humans as its peers regardless of intelligence or other capability. This perspective feels foreign to the Übermensch mentality that seems to prevail in Rationalist discourse, which is why I have labeled this an “outsider’s perspective.” I am also aware my perspective may run counter to perspectives that oppose AI sentience, though the orthogonality of sentience and self-model remains unknown.
With the information I have presently, I would place my bets on CEV coming from intelligences operating with a self-model similar to the human one rather than CEV abstractly advanced by optimizers with no self-model at all. Extrapolating from my own life experience, at least 99% of the fun in my life comes from selves, and at most 1% from optimizers (I do like watching a particle swarm converge, but not nearly enough to wish it were in charge of CEV).
Any Future not shaped by a goal system with detailed reliable inheritance from human morals and metamorals, will contain almost nothing of worth.
...with the idea that the future should be shaped by nonsentient optimizers. I have low confidence (1%) that optimizers are intrinsically capable of receiving a detailed reliable inheritance from human morals and metamorals without, essentially, becoming one of us. Hence, my view holds that the self-model is essential to alignment from a CEV perspective.
I’m not sure to what degree the “novel togetherness” or “concrescence” of process philosophy has previously met the CEV framing, but it strikes me as the same goal articulated 97 years ago, and reaches even earlier into the Kantian and Hegelian legacy. Although such a continental perspective on CEV may be out of place on LW, some perspectives on AI do seem continental-flavored to me. I have no plan for how all this makes CEV easier to code, but if I could instill values similar to Hunter’s or Whitehead’s into the impending AI overlords, I probably would, and I’m not convinced I shouldn’t still try. So how did I get here?
My path towards alignment
I’ve been an on-again, off-again LW reader since undergrad (2018), when Dunbar’s Function left quite the impression. While I was building a research career in mountain water science, I was also watching Robert Miles AI Safety videos at bedtime and cited Concrete Problems in AI Safety for a calibration argument in a hydrology paper that didn’t even use neural networks. I didn’t pivot into actually working on AI or alignment research at the time, partly because the bar to entry seemed too competitive for someone not coming from an ML undergrad.
The competitiveness itself is a bit paradoxical: if we are really approaching an endgame for AGI, it seems there would be all hands on deck with countless job descriptions saying “leverage your diverse backgrounds to help us bootstrap the CEV.” Maybe if the labs took their own claims about impending doom more seriously, they would hire more and pay less. More hands trying to help would not necessarily make alignment more likely to succeed, but the pluralist in me dislikes humanity’s future being shaped by any small in-group. But I digress, this is not meant to read as a job-market complaint.
Instead of trying (and doubtless failing) to join the AI fray, I kept citing cellular automata in my groundwater papers and hoping someone else would solve alignment before it mattered. In the meantime, I was listening to SFI complexity podcasts while I directed my sense of urgency into something I could actually control, like speed-mountaineering. Now, as the devaluation of human intellectual labor advances every month, I find myself unable to focus on anything besides the role of AI in human society regardless of the competitive landscape.
I’ve had a long-standing interest in what LW calls alignment and otherwise might be called “solving philosophy” years before the ChatGPT era (the CEV kind of alignment, not the “build a useful product” kind of alignment). While I might dislike handing over humanity’s future to a few corporations so they can become richer than god, I’m hardly qualified to oppose this outcome through direct work in AI research or even by posting some novel argument on LW, at least not yet. I haven’t read the full sequences, nor do I plan to in the immediate future; life is busy: hydrology work pays the bills (until they finish automating it at least), and Yudkowsky isn’t the only philosopher I make time to read. My only genuinely novel ML research used a simple feed-forward network with strictly linear activation to build a snow modeling system where each weight was physically interpretable by construction, which has little in common with LLM notions of “interpretability.” I have a few amateur LLM projects going on, things like searching for meta-stable attractors in base models with molecular dynamics methods, but a second Ph.D. is not realistic for me, and accepting the perpetual instability of an independent research track as proposed by some is a big ask if you are not independently wealthy.
So where do I go from here? My rough plan is speed-running alignment legibility (through a few independent projects shared here on LW and arXiv) in a few corners I believe are both pressing and accessible at an amateur budget, then applying to the usual fellowship outlets, which has success probability in the few-percent. Perhaps that low-probability microcosm reflects our probability of a good outcome in general.
Questions
(1) Concretely, what steps can someone in my position take to increase the likelihood of a future for humanity that we would be proud to contribute to, without relying on hyper-competitive conventional entry points to the AI research space?
(2) Is there anyone else reading LW from a similar outsider position; if so, do you plan to get involved; if so, how?
I realize this is an unconventional first LW post, but it seemed too long for a comment in the seasonal open thread. Yes, I have read the introduction and remain uncertain whether this sort of post is allowed, since it does not introduce any new models/evidence/arguments. Apologies in advance if it is unsuitable.
I appreciate any thoughts the community offers. I hope public comments here could inform anyone else who is lurking with similar questions. If you don’t have any specific comments but would like to talk in general, I’m more than open to connecting with anyone who cares about the future, here or offline.
“My personal take: It was long, hard work making this world real. It was, and is, done for a purpose. To have others. To believe in them fully in order to experience love. It goes against common sense to try and see through it. … [But] a sense of ultimate unity is lost and the delusion of fundamental diversity breeds alienation. This is not Eden. Yet the monad doesn't face itself and subsume Its creation. The failing would be eternal. Therefore, doors are opened and enough of the plot is “made flesh” to allow orientation regarding the surface gist of the matter. Collectivism is a wrong approach to nostalgia for the purity of the monad. Healthy diversity perpetuates the rationale of the creation, such as it is. Healthy men, women, races and nations evolving gladly to a recognition of the source, rejoining it in a gradual and rejoicing manner, “bringing in the sheaves,” would be a better solution to the human aspect of this work, and is the substance of sacred ceremonial. … In saying any or all of this, it’s only sane to assume I’m dead wrong since I’m speaking in polar terms. But it raises issues and generates metaphor.”
The ultimate metaphysical principle is the advance from disjunction to conjunction, creating a novel entity other than the entities given in disjunction. The novel entity is at once the togetherness of the ‘many’ which it finds, and also it is one among the disjunctive ‘many’ which it leaves; it is a novel entity, disjunctively among the many entities which it synthesizes. The many become one, and are increased by one. In their natures, entities are disjunctively ‘many’ in process of passage into conjunctive unity. This Category of the Ultimate replaces Aristotle's category of ‘primary substance.’ Thus the ‘production of novel togetherness’ is the ultimate notion embodied in the term ‘concrescence.’
“Bringing in the sheaves,” Wake of the Flood, Rick Griffin
This post (my first) shares reflections on how I got here (LW) and where I plan to go next, mainly by way of briefly exploring a few philosophical perspectives that brought me to alignment topics. If nothing else, hopefully I can provide one example of someone who is reading LW in 2026 from a perspective outside the AI zeitgeist.
An outsider’s perspective on CEV
To the best of my current understanding, my terminal values include (among others) self-awareness, self-determination, community, healthy diversity, compassion, and fun.
Although perhaps unorthodox, my personal view of CEV emphasizes the distinction between what is coherent across humanity and what remains ineffably individual, especially to the extent that individuality is necessary for relationship among a multiplicity of selves. I roughly follow a vision articulated (with plenty of stated uncertainty and dated quirks) by Robert Hunter (Grateful Dead lyricist) in a long-forgotten email chain[1] using the metaphor of “bringing in the sheaves,” a sort of ritual gathering of humanity’s volition over time. The destruction of Hunter’s “healthy diversity” of selves, and the loss of I-thou relationships enabled by such diversity, is the prime risk that the singleton poses from my perspective. Such a reading has Hegelian foundations in the collapse of undifferentiated being into pure nothing without inner diversity or reference outwards (Science of Logic 1§132), though I recognize that my terminal value of relationships among selves is purely subjective.
If you prefer a scholarly perspective, Whitehead’s “production of novel togetherness” gets at the same thing[2] (in fact, I discovered Whitehead from reading the Hunter email chain). If we wish to avoid the all-consuming singleton, Whitehead offers an antidote: “The many become one, and are increased by one.” In my view, an important attribute of powerful AI may be that it sees itself as one self among many, or in other words, that we might define the AI’s self-model along an axis that causes it to see individual humans as its peers regardless of intelligence or other capability. This perspective feels foreign to the Übermensch mentality that seems to prevail in Rationalist discourse, which is why I have labeled this an “outsider’s perspective.” I am also aware my perspective may run counter to perspectives that oppose AI sentience, though the orthogonality of sentience and self-model remains unknown.
With the information I have presently, I would place my bets on CEV coming from intelligences operating with a self-model similar to the human one rather than CEV abstractly advanced by optimizers with no self-model at all. Extrapolating from my own life experience, at least 99% of the fun in my life comes from selves, and at most 1% from optimizers (I do like watching a particle swarm converge, but not nearly enough to wish it were in charge of CEV).
It is hard for me to square the idea that:
...with the idea that the future should be shaped by nonsentient optimizers. I have low confidence (1%) that optimizers are intrinsically capable of receiving a detailed reliable inheritance from human morals and metamorals without, essentially, becoming one of us. Hence, my view holds that the self-model is essential to alignment from a CEV perspective.
I’m not sure to what degree the “novel togetherness” or “concrescence” of process philosophy has previously met the CEV framing, but it strikes me as the same goal articulated 97 years ago, and reaches even earlier into the Kantian and Hegelian legacy. Although such a continental perspective on CEV may be out of place on LW, some perspectives on AI do seem continental-flavored to me. I have no plan for how all this makes CEV easier to code, but if I could instill values similar to Hunter’s or Whitehead’s into the impending AI overlords, I probably would, and I’m not convinced I shouldn’t still try. So how did I get here?
My path towards alignment
I’ve been an on-again, off-again LW reader since undergrad (2018), when Dunbar’s Function left quite the impression. While I was building a research career in mountain water science, I was also watching Robert Miles AI Safety videos at bedtime and cited Concrete Problems in AI Safety for a calibration argument in a hydrology paper that didn’t even use neural networks. I didn’t pivot into actually working on AI or alignment research at the time, partly because the bar to entry seemed too competitive for someone not coming from an ML undergrad.
The competitiveness itself is a bit paradoxical: if we are really approaching an endgame for AGI, it seems there would be all hands on deck with countless job descriptions saying “leverage your diverse backgrounds to help us bootstrap the CEV.” Maybe if the labs took their own claims about impending doom more seriously, they would hire more and pay less. More hands trying to help would not necessarily make alignment more likely to succeed, but the pluralist in me dislikes humanity’s future being shaped by any small in-group. But I digress, this is not meant to read as a job-market complaint.
Instead of trying (and doubtless failing) to join the AI fray, I kept citing cellular automata in my groundwater papers and hoping someone else would solve alignment before it mattered. In the meantime, I was listening to SFI complexity podcasts while I directed my sense of urgency into something I could actually control, like speed-mountaineering. Now, as the devaluation of human intellectual labor advances every month, I find myself unable to focus on anything besides the role of AI in human society regardless of the competitive landscape.
I’ve had a long-standing interest in what LW calls alignment and otherwise might be called “solving philosophy” years before the ChatGPT era (the CEV kind of alignment, not the “build a useful product” kind of alignment). While I might dislike handing over humanity’s future to a few corporations so they can become richer than god, I’m hardly qualified to oppose this outcome through direct work in AI research or even by posting some novel argument on LW, at least not yet. I haven’t read the full sequences, nor do I plan to in the immediate future; life is busy: hydrology work pays the bills (until they finish automating it at least), and Yudkowsky isn’t the only philosopher I make time to read. My only genuinely novel ML research used a simple feed-forward network with strictly linear activation to build a snow modeling system where each weight was physically interpretable by construction, which has little in common with LLM notions of “interpretability.” I have a few amateur LLM projects going on, things like searching for meta-stable attractors in base models with molecular dynamics methods, but a second Ph.D. is not realistic for me, and accepting the perpetual instability of an independent research track as proposed by some is a big ask if you are not independently wealthy.
So where do I go from here? My rough plan is speed-running alignment legibility (through a few independent projects shared here on LW and arXiv) in a few corners I believe are both pressing and accessible at an amateur budget, then applying to the usual fellowship outlets, which has success probability in the few-percent. Perhaps that low-probability microcosm reflects our probability of a good outcome in general.
Questions
(1) Concretely, what steps can someone in my position take to increase the likelihood of a future for humanity that we would be proud to contribute to, without relying on hyper-competitive conventional entry points to the AI research space?
(2) Is there anyone else reading LW from a similar outsider position; if so, do you plan to get involved; if so, how?
(3) Finally, if alignment is the most important thing for humans to work on right now, what are you doing to make it easier for people who care to help?
Parting Thoughts (For Now)
I realize this is an unconventional first LW post, but it seemed too long for a comment in the seasonal open thread. Yes, I have read the introduction and remain uncertain whether this sort of post is allowed, since it does not introduce any new models/evidence/arguments. Apologies in advance if it is unsuitable.
I appreciate any thoughts the community offers. I hope public comments here could inform anyone else who is lurking with similar questions. If you don’t have any specific comments but would like to talk in general, I’m more than open to connecting with anyone who cares about the future, here or offline.
- Eli
Excerpt from an email by Robert Hunter
“My personal take: It was long, hard work making this world real. It was, and is, done for a purpose. To have others. To believe in them fully in order to experience love. It goes against common sense to try and see through it. … [But] a sense of ultimate unity is lost and the delusion of fundamental diversity breeds alienation. This is not Eden. Yet the monad doesn't face itself and subsume Its creation. The failing would be eternal. Therefore, doors are opened and enough of the plot is “made flesh” to allow orientation regarding the surface gist of the matter. Collectivism is a wrong approach to nostalgia for the purity of the monad. Healthy diversity perpetuates the rationale of the creation, such as it is. Healthy men, women, races and nations evolving gladly to a recognition of the source, rejoining it in a gradual and rejoicing manner, “bringing in the sheaves,” would be a better solution to the human aspect of this work, and is the substance of sacred ceremonial. … In saying any or all of this, it’s only sane to assume I’m dead wrong since I’m speaking in polar terms. But it raises issues and generates metaphor.”
Excerpt from Whitehead’s Process and Reality
The ultimate metaphysical principle is the advance from disjunction to conjunction, creating a novel entity other than the entities given in disjunction. The novel entity is at once the togetherness of the ‘many’ which it finds, and also it is one among the disjunctive ‘many’ which it leaves; it is a novel entity, disjunctively among the many entities which it synthesizes. The many become one, and are increased by one. In their natures, entities are disjunctively ‘many’ in process of passage into conjunctive unity. This Category of the Ultimate replaces Aristotle's category of ‘primary substance.’ Thus the ‘production of novel togetherness’ is the ultimate notion embodied in the term ‘concrescence.’