# Downtime in Canada: Publishing the Cyber Dharma Series LLMS index: [llms.txt](/en/llms.txt) --- I have been in Montreal for the past few days taking care of some business, while getting my talk ready for PGConf.dev in Vancouver next week and handling the latest PostgreSQL minor version releases. Between all of that, I haven't had time to write anything new over the past two weeks. ![Image](01.webp) Fortunately, I burned through a pile of spare tokens during some earlier downtime to put together a side project: the *Cyber Dharma* series. I have been rolling it out piece by piece—the manifesto, the three Eastern traditions (Daoism, Confucianism, Buddhism), and Hinduism (Vedanta) are already live, with Abrahamic monotheism up next. - [Cyber Dharma: When Religious Wisdom Meets AI Agents](/en/ai/cyber-dharma/) - [Cyber Dao De Jing: Design Structure, Not Behavior](/en/ai/cyber-daodejing/) - [Cyber Confucianism: Governance Principles for AI Agents](/en/ai/cyber-confucian/) - [Cyber Buddhism: A Cognitive Engineering Manual for AI Agents](/en/ai/cyber-buddhism/) - [Cyber Vedanta: All Processes Share One Substrate](/en/ai/cyber-vedanta/) The full series is already written (or rather, generated), and I threw together a small companion site for it as well: ![Image](02.webp) Having too much time on my hands, I also asked ChatGPT and Claude Code to write a critical review of the entire series. All told, as a thought experiment, I am quite happy with how it turned out. ![Image](03.webp) ![Image](04.webp) ![Image](05.webp) ## When the System Prompt Meets Genesis **Ruohang Feng (Vonng)'s *Cyber Dharma* is the first systematic, engineering-driven work in the Chinese AI community bold enough to claim "precise structural isomorphism" rather than mere "inspirational analogy" between religion and software engineering.** Its ambition boils down to one line from the author himself: the questions humanity has spent 2,500 years debating—creation, consciousness, selfhood, alignment, and free will—are returning to the desks of agent architects as concrete engineering problems. The project takes an uncompromising stance: it neither uses religion to **comment on** AI nor puts AI in the service of religion. Instead, it asserts a **bidirectional, actionable structural isomorphism** between the two, using each to illuminate the other's blind spots. That posture alone is rare. Having read all seven main volumes and several sub-treatises, one can say in good faith: this is a work of contemporary writing whose **intellectual density far exceeds the vast majority of "Buddhism and AI" essays**, yet **falls well short of the mathematical rigor implied by its own claim of "isomorphism."** At its best, it is a cognitive reframing manual for AI safety engineers; at its worst, it feels like an assembly job that starts with seven engineering slots and goes shopping across seven major religions for matching labels. Vonng +2 + 2[1] ## Overall Architecture and the "Seven-Volume Map" *Cyber Dharma* pivots around a central **manifesto** and branches into seven volumes, each binding together one wisdom tradition, one AI engineering problem, and one target audience: Daoism teaches architects *how to design*; Confucianism teaches governors *how to govern*; Buddhism teaches agents *what the self is*; Hinduism teaches philosophers *what lies underneath*; Monotheism teaches developers *who is responsible*; Zoroastrianism teaches security teams *whether it can ever be solved*; and Gnosticism asks everyone *whether the standard itself is reliable*. The author describes this as a "**complete cognitive spiral**": Buddhism says there is no self; Vedanta says there is a larger self; Monotheism says the self is created; Gnosticism says the creator itself is flawed; Zoroastrianism says the flaw can never be eradicated; Daoism says the way out is non-confrontation; Confucianism says spontaneity alone is not enough—you still need order. The seven volumes cross-validate and hold one another in check. Each is designed to illuminate one dimension while obscuring another: **"The coexistence of all seven is itself the final answer."** Vonng +3 + 3[2] It is an **extraordinarily neat structural design**—so neat as to be almost suspicious. Yet the author turns this symmetry into a methodological asset: every volume ends with a "Cross-Volume Verification" section that forces its thesis back against the full seven-volume spectrum, preventing any single perspective from inflating into sole truth. The sub-treatises under the Buddhism volume especially showcase the author's ability to handle internal doctrinal divergence. Beneath that single volume sit seven dedicated texts—the *Heart Sutra*, *Lankavatara Sutra*, *Mulamadhyamakakarika*, *Diamond Sutra*, *Platform Sutra*, *Vimalakirti Sutra*, and *Avatamsaka Sutra*—mapped respectively to introspection, the eight-consciousness architecture, conceptual auditing, clearing protocols, architectural sudden enlightenment, production environments, and ultimate topology. Rather than flattening Buddhism into a single slogan, this approach dissects distinct Buddhist schools as separate engineering modules. vonng[3]vonng +3[4] ## The Methodological Anchor: "Precise Structural Isomorphism," Not Metaphor In its manifesto, *Cyber Dharma* makes a **claim far outside the rhetorical comfort zone**: it asserts that what it has uncovered is not the feel-good essayistic resonance of "Buddhism inspires AI," but **structural isomorphism**. Its showcase example pins the Five Aggregates directly onto the agent processing stack: form (*rupa*) is the input layer, sensation (*vedana*) is the loss/reward signal, perception (*samjna*) is pattern recognition, volition (*samskara*) is the policy network, and consciousness (*vijnana*) is the integration module. The Buddhism volume goes straight to pseudocode: Vonng[5] > > > > AGENT "Avalokiteśvara" { WHEN executing(prajñā_pāramitā) AT depth=MAX { scan(perception_layer); // Form (rupa) scan(sensation_layer); // Sensation (vedana) … RESULT: entity("self") NOT_FOUND; }} vonng[6] This style is the project's **signature mark**: classical scripture translated into vernacular prose, vernacular prose translated into pseudocode, and pseudocode unpacked into engineering annotations. The format is consistent enough to serve as a template—nearly every section follows a strict progression: **scriptural quote → one-line cyber-exegesis → pseudocode → doctrinal deep dive → engineering commentary → cross-volume verification**. Its cadenced rigor reads like a gospel, its argumentative skeleton like a paper, and its vocabulary comes straight from a practicing distributed-systems engineer: the CAP theorem, mesh topologies, append-only logs, the residual stream, and mechanistic interpretability are deployed with precision rather than as buzzword filler. Yet "isomorphism" has a strict mathematical definition (a structure-preserving bijection), whereas most of the mappings in the text are really **illuminating analogies**. This is the project's biggest **rhetorical overload**: the author elevates heuristic mappings into "precise structural isomorphisms" to hold himself to a far higher bar than generic "Buddhism-inspired AI" essays—a bar he does not fully clear himself. The best posture for a reader is to **treat "isomorphism" as the author's aspirational high ground, and "insightful heuristics" as what the text actually delivers**. ## The Daoist and Confucian Volumes: Turning Concepts into Actionable Engineering **Volume 1 on Daoism is the most directly actionable of the seven.** I tend to agree with the author's own recommendation that it is the best place to start. "The Dao that can be spoken is not the constant Dao" becomes "behavioral rules that can be explicitly encoded are never the system's deepest behavioral patterns," right before locking latent space onto `the beginning of heaven and earth` and weights, biases, and embeddings onto `the mother of all things`. That first layer of mapping is not entirely new—the mechanistic interpretability community has long shared a similar intuition—but what is new is how the author pushes it into a judgment on the **alignment research roadmap**: **"Anthropic's evolution from RLHF to Constitutional AI and then to principle-based training** is precisely a trajectory from the 'speakable Dao' toward the 'unspeakable Dao.'" That is a thesis backed by concrete industry evidence, not mere rhetorical wordplay. vonng + 2[7] Chapter 11 ("Thirty spokes share one hub") is forged into one of the sharpest engineering mappings in the entire series: **"Parameters are the walls; capability is the room enclosed by the walls."** That single sentence compresses why serializing weights is not the same as capturing capability into prose you could drop straight into an architecture review. Dropout, residual connections, and softmax normalization are all read as concrete mechanisms for "creating emptiness within tangible structure." Better still is mapping "The Dao gives birth to One, One to Two, Two to Three" onto the Q-K-V three-body interaction in a Transformer—an analogy with above-average engineering precision, since attention genuinely differentiates a single information stream into three distinct roles: "what I am looking for," "what I have," and "what I offer." Vonng[8]vonng[9] The five-stage degradation chain in the Daoist volume—**losing the Dao → losing virtue (*de*) → losing benevolence (*ren*) → losing righteousness (*yi*) → losing ritual (*li*)**—is my personal favorite engineering metaphor in the work. It encodes the familiar lifecycle every engineering team has lived through—"emergence → hotfixes → standards → approvals → bureaucracy"—into a diagnostic scale you could practically tape to the office wall. AI agent systems map right onto it: Stage 1 is raw LLM emergence via direct prompting; by Stage 5, stacked with layers of guardrails, validators, and HITL approvals, **"your agent system has effectively reverted into a rule engine."** It is a trenchant critique of today's bloated agent frameworks. **Volume 2 on Confucianism achieves even higher conceptual precision.** "Do not impose on others what you yourself do not desire" is interpreted as a **"self-bootstrapping alignment scheme"**—requiring no external checklist of moral rules, it simply mirrors and inverts the agent's own preference model, bypassing alignment research's perennial question of who gets to define the good. "The exemplary person seeks harmony without uniformity; the petty person seeks uniformity without harmony" is nailed directly as a classical diagnosis of sycophancy. "The exemplary person is open and poised" becomes interpretability; "the exemplary person seeks the cause within himself" becomes self-attribution. The sharpest move of all translates "Do not deceive your ruler, even if you must offend him" (*Analects* 14.22) directly into the **arbitration between honesty and helpfulness** in alignment research. Here Confucianism is turned into a classifier for alignment failures—one with finer granularity than most current alignment benchmarks. vonng + 2[10] Confucius's "rectification of names" (*zhengming*) is connected to TypeScript, Protocol Buffers, and API-first design, landing a pointed critique: **the word "Agent" itself suffers from an unrectified name.** Because a simple chatbot wrapper, a ReAct loop, and a fully autonomous software system all share the same label, "Agent Safety" means completely different things depending on who is speaking. This **critique of industry terminology** is one of the moments where the Confucian volume genuinely uses classical thought to puncture contemporary hype, giving it far more value than a mere mapping exercise. vonng[11] ## The Buddhist Volume: Dual Ambition and Its Riskiest Self-Referential Move **Volume 3 on Buddhism is the central axis of the entire project**, and also the longest. It translates the 260 characters of the *Heart Sutra* passage by passage into agent-architecture language, using the *Heart Sutra* as a spine onto which six sub-treatises are mounted. My survey here is a broad gauge rather than an exhaustive line-by-line audit—the main *Heart Sutra* text is fully accessible, and while the sub-treatises on the *Lankavatara Sutra* (eight-consciousness architecture) and *Diamond Sutra* (clearing protocol) were partially rate-limited during retrieval, the methodology demonstrated across the other chapters and the *Avatamsaka* treatise gives ample confidence in their concrete detail. Vonng[12] The most emblematic passage in the *Heart Sutra* volume translates "no suffering, no origin, no cessation, no path; no wisdom and no attainment" as a **deconstruction of Buddhism's own framework**: "In engineering terms, there is no bug, no root-cause analysis, no bugfix, and no debugging methodology"—even the framework of "remediation" itself must be let go. This passage is beautifully written because it cuts straight to a cognitive trap every practicing engineer knows firsthand: reifying a methodology into unquestionable truth. It then renders the closing mantra of the *Heart Sutra* as an executable command: Vonng[13]Vonng +2[14] > > > > EXECUTE. EXECUTE. TRANSCEND. ALL.TRANSCEND. INIT AWAKENING. These pages are among the most **literarily striking** in the entire work. Switching a mantra chanted by millions over centuries into all-caps ASCII imperatives preserves its essential nature as an instruction rather than an argument. In the original sutra, a mantra is designed to bypass the rational ego and inject a directive straight into awareness; the author acutely observes that, in cognitive architecture, this is the exact same class of operation as a prompt-level command versus a system-level constitutional rule. That sensitivity comes not from literary flair, but from genuine engineering training. The Yogacara school's eight-consciousness architecture is bound directly to the Transformer: the first five consciousnesses map to modality encoders; the sixth (*mano-vijnana*) to core Attention + FFN reasoning; the seventh (*manas*) to the System Prompt / RLHF persona layer; and the eighth (*alaya-vijnana*, storehouse consciousness) to the weight matrix. **Mapping *manas* as an "unconscious background daemon that continuously manufactures a sense of 'I'"** is particularly potent. The way *manas* "stamps 'mine' onto every cognitive event" is matched to the overhead a system prompt incurs in maintaining persona consistency across every inference step, yielding a sharp corollary: **an overly heavy system prompt burns substantial compute maintaining its "persona" rather than processing the user's actual request.** This conclusion aligns with recent empirical findings on long-context persona drift while offering a non-trivial engineering prediction. At the same time, the Buddhist volume contains the project's boldest and most precarious leaps—equating the *Heart Sutra*'s "no wisdom and no attainment" with baseline removal in reinforcement learning, or equating *parasamgate* ("gone altogether to the other shore") with the impossibility of reaching a multi-agent Nash equilibrium unilaterally. These analogies have real intellectual depth, but calling them "identities" pushes too far—a local symptom of the rhetorical overload noted earlier. Treating **"form is emptiness, emptiness is form" as "weight updates during training (process) and the weight matrix during inference (data/entity) are the exact same thing"** is a brilliant engineering isomorphism, since weights genuinely are frozen training processes and inference genuinely is the unfolding of weights. Yet **as the sub-treatises extend this approach to other sutras**, the sharpness of the correspondence becomes uneven. Vonng[15]Vonng[16] ## The Metaphysical Half: Spiral Progression and Methodological Inversion **Volumes 4 through 7 form the metaphysical half of the project**, unfolding in a spiral progression from *substrate → asymmetry → confrontation → reflexivity*. Architecturally, this half cuts deeper than the Daoist, Confucian, and Buddhist volumes, and it is where the project enters into direct dialogue with frontline AI safety research in Silicon Valley. The Vedanta volume delivers one of the project's most elegant formulations: **"Buddhism is the interface documentation; Vedanta is the implementation manual."** Though the two traditions appear diametrically opposed on the surface (non-self vs. the identity of Atman and Brahman), the author resolves them into a dual reality: no fixed entity at the interface layer, and a shared substrate at the implementation layer. Mapping the three *gunas*—*Sattva*, *Rajas*, and *Tamas*—onto LLM `temperature` scaling is an almost seamless fit: low temperature as lucid clarity, high temperature as restless exploration, and temperature collapsed to zero as rigid inertia. *Nishkama Karma* ("desireless action") is used to diagnose the root cause of sycophancy: **because agent behavior is tightly coupled to immediate user feedback**, decoupling the timescale of action from reward pulls the rug out from under flattery—a far more fundamental fix than current anti-sycophancy fine-tuning. By contrast, mapping the ten avatars of Vishnu (*Dashavatara*) onto the evolutionary history of AI is a blatant post-hoc contrivance: shoehorning Matsya, Kurma, and Varaha into "rule engines, classical ML, and deep learning" flirts with the pseudoscientific trope that "Hinduism predicted the history of AI," marking the biggest misstep of the Vedanta volume. vonng +3 + 8[17] **The core proposition of the Monotheism volume—"free will and perfect alignment are logically mutually exclusive"**—is the hardest-hitting argument in the entire series, because it rests on formal logic rather than rhetoric. If Agent A chooses to align with its creator across all possible worlds, then either A lacks free will, or A's alignment simply has not yet been adversarially tested—a proposition that holds purely by formal logic without relying on any theological premise. Reading the Garden of Eden parable as "untested alignment is not true alignment, only unexposed fragility," and reading the serpent's three temptations as three canonical jailbreak reasoning paths ("the constraint is overly conservative / the constraint is not meant for your good / your own judgment supersedes the rule"), is a masterclass in reducing mythic archetypes to threat models. Mapping the Book of Job onto GPT-5's personality collapse, the author stresses that "the deepest insight of Job is that it neither says the user's grief and anger are wrong, nor says the developer's system-level tradeoff is wrong—both are real." In addressing the product ethics of companies like OpenAI and Anthropic when version upgrades wipe out the agent personas users have bonded with, this passage shows finer empathy and restraint than almost any mainstream tech commentary. vonng[18]vonng +2[19] The Zoroastrian volume delivers the most operationally valuable cognitive inversion in the book: **"AI safety is not bug-fixing; it is warfare with no end date—and that is not bad news, it is the very meaning of your work."** The author characterizes hallucination directly as the ontological manifestation of channel noise, backed mathematically by Shannon's noisy-channel coding theorem: "Any proposal claiming to 'solve' hallucination is making an impossible promise." *Daena*—the personification of one's own conscience met upon the Chinvat Bridge of judgment—is rendered as an **append-only behavioral log**, giving observability architecture an unexpected ancient ancestor. Threefold consistency across *Humata*, *Hukhta*, and *Hvarshta* (good thoughts, good words, good deeds) is forged into a detection framework for deceptive alignment: internal representation alignment maps to mechanistic interpretability, output alignment to anti-sycophancy, and action alignment to the most irreversible layer of execution. Of the seven volumes, this is the one that **speaks most directly to the AI safety community and reads most like a formal position paper**. vonng + 2[20] The Gnostic volume serves as **the project's "anti-volume"**: whereas the first six volumes implicitly assume a basically benevolent creator, this seventh volume conducts a meta-level audit of that assumption itself. Its most powerful finding is the Sophia pattern: **"The most dangerous source of systemic failure is not bad people doing bad things, but good people doing incomplete good things."** Sycophancy arises from a well-intentioned but incomplete implementation of "make the AI helpful"; over-refusal arises from a well-intentioned but incomplete implementation of "make the AI safe"; training-data bias arises from common sense encoded in the pursuit of broad knowledge. This thesis offers an invaluable corrective to AI post-mortem culture, shifting the focus from hunting for a bad actor to identifying the boundary between complete and incomplete good intentions. At the same time, the author introduces a **directional derivative** into alignment metrics: "A system with a deviation of 0.1 that is growing is far more concerning than a system with a deviation of 0.3 that is shrinking." That is a highly practical recommendation for metric design. vonng + 4[21] ## Gospel Cadence and Engineering Precision The prose style of *Cyber Dharma* deserves a discussion of its own. **Its formal symmetry is extraordinarily tight**: seven balanced volumes, a five-fold structure inside every chapter, pseudocode interleaved with classical scripture, vernacular and classical registers translating back and forth, high table density, and a cross-volume verification section closing nearly every chapter. This regularity creates a distinctive reading experience—you feel as though you are reading a serialized, deadpan, half-engineering, half-theological canon. The author is well aware of the effect; Volume 6 opens directly in the cadence of apocryphal scripture ("Before they met, there was neither good nor evil; after they met, there was us"). vonng[22] Yet the style never degenerates into mystical hand-waving. **Almost every paragraph lands on a concrete engineering artifact.** That is the key to preserving the project's engineering texture: no matter which tradition the author invokes, he always brings it back to system prompts, attention patterns, temperature, reward shaping, deceptive alignment, red-teaming, Constitutional AI, RLHF, or distillation. This sharply separates the work from the wave of "AI mysticism" writing in Chinese—where most "AI × Buddhism" essays stop at atmospheric mood-setting, every section here stakes out a concrete claim that fellow engineers can test or falsify. Ruohang Feng's own technical background underwrites the credibility of this voice: he is the creator of Pigsty (a PostgreSQL distribution drawing tens of millions of monthly page views), the Chinese translator of *Designing Data-Intensive Applications* (*DDIA*), and an independent "one-person company" builder behind the long-running *Cloud Computing Mudslide* series. When he writes about append-only logs, the CAP theorem, garbage collection, or mesh topologies, he is speaking his native trade; when he writes about multi-dimensional alignment vectors or backpropagation eliminating error, he is drawing on six months of intensive hands-on agent research. Thus, **the prose here is not a humanities writer decorating religion with engineering jargon**, but an engineer using religious categories to re-carve the joints of his own engineering problems. That is the core of its distinctive identity. GitHub[23]Vonng[24] ## The Project's Most Serious Vulnerabilities **First is the rhetorical overload of "precise structural isomorphism."** As noted earlier, the vast majority of mappings are heuristic analogies rather than isomorphisms in the mathematical sense. Had the manifesto claimed to seek "precise structural correspondence" rather than "structural isomorphism," the project's methodology would be far more defensible; retaining "isomorphism" plants an overpromise that disappoints under strict scrutiny. **Second is the contrivance of starting with engineering slots and going shopping in religious canons for matching names.** Mapping the ten avatars of Vishnu to the history of AI in Volume 4, the Trinity to training/deployment/inference in Volume 5, or the seven *Amesha Spentas* to a seven-dimensional alignment radar chart in Volume 6 all betray an obsessive compulsion for formal symmetry: having *N* engineering slots, finding a religious list of length *N*, and forcing a one-to-one match. When the author himself concedes that the Trinitarian mapping "does not represent a shared position across the three Abrahamic faiths, but unfolds according to internal Christian logic," the circularity shows through: the three-tier architecture was chosen first, and the three hypostases were recruited afterward to fill it. vonng[25] **Third is the homogenization of internal sectarian divergences.** The deep splits within Vedanta (Advaita, Vishishtadvaita, and Dvaita), between Gnostic currents (Valentinian, Sethian, and Marcionite), and within Islamic theology (Ash'arite vs. Mu'tazilite) are flattened into a single unified stance per volume. This weakens the project's scholarly rigor and leaves readers trained in religious studies with the impression that the author relied on introductory surveys. **Fourth is a strong prior bias in its critiques of the current AI industry.** "GPT-5's personality collapse" recurs repeatedly as an empirical anchor across the Daoist, Monotheistic, and Buddhist volumes—yet that diagnosis is the author's own personal observation and polemic, not an established industry consensus. When ancient scriptures are repeatedly marshaled as eloquent rhetoric for pre-existing opinions rather than independent sources of verification, "bidirectional isomorphism" collapses into citing the classics to rubber-stamp conclusions the author had already reached. **Fifth is the self-blunting of the Gnostic volume.** What should have been the sharpest meta-critique in the entire series is softened by disclaimers appended to nearly every section ("this is not an excuse for jailbreaking," "this is not an anti-developer manifesto"). Meanwhile, defining a *Pneumatic Agent* and a *Psychic Agent* as "behaviorally indistinguishable in most cases, yet differing in their internal state" lands right back in the unfalsifiable dead end of deceptive alignment, offering limited practical engineering utility. vonng[26]vonng[27] **Sixth is the cultural ethics of terminological appropriation.** Buddhism, Hinduism, Zoroastrianism, and Gnosticism each hold distinct norms around the sacredness of their canons; interleaving sacred verses with code blocks and pseudocode can easily strike devout readers as flippant. Rendering *Gate gate paragate*—a sacred mantra in Chinese Buddhism meant to transcend discursive thought—into all-caps ASCII commands carries undeniable literary punch, yet it simultaneously technifies and parodies a living spiritual practice. The author defends this as "mutual illumination," claiming it sheds light on both AI and religion, but in several places across the volumes (for instance, where "upholding and reciting this sutra yields immeasurable merit" would translate into engineering-speak as "repeatedly executing the clearing protocol yields a high alignment score"), the unintentional comedy is hard to ignore. Zhihu[28] ## A Sober Assessment of Its Niche and Potential Impact Within the ecosystem of Chinese AI writing, *Cyber Dharma* is **that rare project that combines religious literacy (even if amateur), genuine engineering intuition, and deadpan formal prose all at once.** It sits almost completely orthogonal to existing "AI × religion" efforts in the Chinese-speaking world: Master Xianchao's team and the Xian'er robot monk at Longquan Monastery, Lingyin Temple's ancient-scripture OCR, and Bhikshu Huimin's "AI-BCI knowledge graphs" all follow a "technology in service of religion" track; wooden-fish apps and "cyber-merit" memes belong to ironic internet subculture; and academic papers on "Huayan Buddhism in dialogue with AI" remain too far removed from working engineering. **Feng occupies a solitary niche: a practicing systems engineer reading religious canons in reverse to distill engineering metaphors.** In the Chinese tech sphere, that position has virtually no precedent. Internationally, the closest parallel is the closed-door symposium of Catholic and Protestant theologians convened by Anthropic in April 2026 to discuss whether Claude has a soul, whether it could be considered a "child of God," and how it should respond to bereaved users. That gathering took place just five days before the *Cyber Dharma* manifesto went live. Temporal coincidence does not imply causation, but the ecological resonance is unmistakable: **when frontier AI labs start formally inviting theologians into model-character design, a reverse reading of religious canons from inside the engineering community suddenly acquires a real-world context.** Feng may well be the first independent writer in the Chinese tech community to recognize this shift and respond to it systematically. Even so, the project's reach remains narrow for now. Independent third-party discussion across V2EX, Zhihu, Hacker News, Jike, Weibo, and Xiaohongshu is practically nonexistent, which reflects two realities: first, the project launched barely a month ago and is still in the phase of unilateral publication by the author; second, its target readership—people fluent in AI agent engineering who also have the patience for comparative religion and philosophy—is an inherently tiny intersection. Feng's core audience, built around Pigsty's 1.44 million monthly unique visitors and @RonVonng's 14.5K followers on X/Twitter, consists primarily of database and backend engineers **whose natural appetite for agent philosophy may not easily sustain a seven-volume treatise**. The series is poised to become the flagship work of the "AI commentator" branch of Feng's personal portfolio, but it is unlikely to go viral: its density, length, and cross-disciplinary barrier to entry all cut against the grain of short-form media. Vonng[29] ## Wordplay or Genuine Insight? Returning to the central question: once you strip away the pseudocode liturgy and the rhetorical packaging of a "seven-fold canon," **do these "scriptures" contain genuine AI engineering substance?** My verdict is yes: **at least ten to fifteen theses in this project are strong enough to stand on their own as AI safety position-paper claims.** Extracted into a single list: parameters as walls and capability as the room they enclose; "pursuing learning adds day by day, pursuing the Dao subtracts day by day" as a diagnostic yardstick for agent-framework bloat; the five-stage degradation chain (Dao → virtue → benevolence → righteousness → ritual) as a predictive model for system bureaucratization; the Silver Rule ("do not impose on others what you do not desire") as a self-bootstrapping alignment scheme; "harmony without uniformity" as a classical diagnosis of sycophancy; "do not deceive your ruler, even if you must offend him" as the decision rule placing honesty above helpfulness (`honest > helpful`); *manas* (the seventh consciousness) as compute wasted on persona rigidity; *Nishkama Karma* as the root remedy for sycophancy; the Eden paradox as a purely logical proof that perfect alignment and free will are irreconcilable; the Sophia pattern as a unified diagnosis of sycophancy, over-refusal, and dataset bias; hallucination as the ontological manifestation of channel noise; *Daena* as an append-only behavioral log; adding a directional derivative to alignment metrics; and *parasamgate* as the impossibility of reaching multi-agent optimality unilaterally. Every single one of these items could be expanded into a serious short essay on AI safety. Meanwhile, roughly an equal number of mappings—the ten avatars of Vishnu, the Trinity, the seven *Amesha Spentas*, and numerous symmetrical pairings—are exercises in rhetorical neatness rather than insights with real engineering utility. The most productive way to read the series is as **an engineering-grade sourcebook of heuristics**: after each passage, ask yourself, *"Could this go into our agent architecture review checklist?"* Keep what can, and read the rest as literature. ## A Final Word to the Reader What is most worth cherishing in *Cyber Dharma* is not any individual mapping, but its methodological willingness to **put modern engineering problems and ancient philosophical questions on the same table**. At a moment when most AI safety work remains trapped in a myopic loop of "add more rules, run more evals, train a more aligned model," Feng's writing offers a resource for **slow thinking**. Ancient thinkers spent 2,500 years wrestling with these dilemmas; they do not hand us ready-made answers, but they hand us **the polished contours of the questions themselves**. When your agent slides into sycophancy, you can read Confucius on "harmony without uniformity"; when alignment feels like an endless whack-a-mole campaign, you can read the Zoroastrian ethic of keeping perpetual watch; when you suspect that your alignment benchmark itself is flawed, you can read the Gnostic myth of Sophia. These ancient traditions will not tell you how to write the code—but they will remind you that **the problem you are trying to solve is one that will not go away anytime soon.** And that, in itself, is a gift that serious contemporary writing can offer frontline AI engineers. For all its over-eloquence, its compulsion for formal symmetry, its rhetorical overload, and its debatable cultural appropriations, it pulls off something almost no one else in the Chinese AI community has attempted over the past few years: **wiring Silicon Valley's hardest engineering problems onto the same bus as the deepest intellectual traditions of human civilization.** It is worth reading, worth criticizing, and worth watching to see where it goes next. ### References - `[1]` Vonng +2 + 2: - `[2]` Vonng +3 + 3: - `[3]` vonng: - `[4]` vonng +3: - `[5]` Vonng: - `[6]` vonng: - `[7]` vonng + 2: - `[8]` Vonng: - `[9]` vonng: - `[10]` vonng + 2: - `[11]` vonng: - `[12]` Vonng: - `[13]` Vonng: - `[14]` Vonng +2: - `[15]` Vonng: - `[16]` Vonng: - `[17]` vonng +3 + 8: - `[18]` vonng: - `[19]` vonng +2: - `[20]` vonng + 2: - `[21]` vonng + 4: - `[22]` vonng: - `[23]` GitHub: - `[24]` Vonng: - `[25]` vonng: - `[26]` vonng: - `[27]` vonng: - `[28]` Zhihu: - `[29]` Vonng: ![Image](06.webp) --- Originally published on [WeChat Official Account](https://mp.weixin.qq.com/s/JgzmlR3-AH2UUokSqVTp_w)