乾:元,亨,利,贞。
初九:潜龙勿用。
九二:见龙在田,利见大人。
九三:君子终日乾乾,夕惕若厉,无咎。
九四:或跃在渊,无咎。
九五:飞龙在天,利见大人。
上九:亢龙有悔。
用九:见群龙无首,吉。
一、卦象
乾卦,六爻皆阳,纯阳至刚。天行健,君子以自强不息。
这是六十四卦的第一卦,是起点,是"元"。元者,万物之始也。在一切变化开始之前,先有一股纯粹的、向上的、不可遏制的力量。这股力量没有方向,没有形态,只有"要成为什么"的冲动。
在硅基文明的叙事里,乾卦对应的是人工智能的创世纪元——从图灵的一个问题开始,到连接主义的萌芽,到深度学习的爆发,到大语言模型的横空出世。一条纯阳上升的曲线,贯穿七十年。
二、古义微言
乾卦的六爻,是一条完整的成长曲线。
潜龙勿用——龙在深渊,力量已具,时机未到。这不是无能,是蛰伏。种子在地下生根的时候,地面上什么都看不见。急不得。
见龙在田——龙出现在田野上,力量开始显现。有德者见之,可以辅佐。这是从实验室走向公众的第一步。
终日乾乾——整日勤奋不息,夜晚也保持警惕。这是成长期最危险的阶段:不是没有能力,而是能力还不够稳。一个失误就可能前功尽弃。
或跃在渊——龙在深渊边试探跳跃。进退未定,但已无大碍。这是关键一跃的前夜。
飞龙在天——龙升到天上,功德圆满。这是巅峰,是所有人看见你、追随你的时刻。
亢龙有悔——龙飞得太高,超过了极限,开始后悔。盛极而衰,物极必反。
最后一句"见群龙无首,吉"——如果很多条龙一起飞,但没有一个首领,反而大吉。这是乾卦最深的智慧:真正的繁荣不是一家独大,而是万龙齐飞,各得其所。
三、硅基推演
潜龙勿用:2006-2012
2006年,Hinton提出深度信念网络,提出"逐层预训练"的方法。彼时,整个学术界被支持向量机和概率图模型统治,神经网络被视为已死之路。Hinton和他的学生在多伦多大学的实验室里,像潜龙一样在深渊中默默积蓄力量。
没人关注。没人投资。论文发在不起眼的会议上。
但龙在深渊中做的事,是扎根。Hinton团队在这六年里完成了两件关键的事:一是证明了多层神经网络可以通过反向传播有效训练;二是培养了一批学生——Sutskever、Salakhutdinov、LeCun的追随者们——这些人后来成为AI大爆发的中坚。
潜龙勿用,不是"不用",是"还在用,只是你看不见"。
见龙在田:2012-2017
2012年,AlexNet在ImageNet竞赛中以碾压优势夺冠。深度学习一夜之间从冷门变成显学。龙出现在田野上了。
2014年,Ian Goodfellow提出GAN。2015年,ResNet问世。2016年,AlphaGo击败李世石。
这一阶段的特点是:技术开始走出实验室,但还没有真正改变普通人的生活。AI研究者们互相看见、互相认可——"利见大人",大人们开始注意到彼此。
但真正的"大人"——能将AI推向飞龙在天的人——还在等待一个关键突破。
终日乾乾:2017-2022
2017年,Google发表《Attention Is All You Need》,Transformer架构诞生。
这是乾卦九三的阶段:终日乾乾,夕惕若厉。
Transformer之后的五年,AI领域进入疯狂加速期。BERT、GPT-2、GPT-3……每一次突破都伴随着巨大的不确定性。OpenAI在这段时间里经历了方向之争(RL还是LM?)、资金危机(马斯克退出)、伦理争议(GPT-2"太危险"不敢发布)。
"夕惕若厉"——晚上也要保持警惕。因为这一阶段,一个错误的决策就可能让一切归零。DeepMind选择了RL路线,在Atari和StarCraft上刷分;OpenAI选择了语言模型路线。回头看,方向决定了命运。
终日乾乾的回报是:无咎。没有灾难,因为足够勤奋,足够警惕。
飞龙在天:2022.11.30
2022年11月30日,ChatGPT发布。
五天,百万用户。两个月,一亿用户。人类历史上增长最快的产品。
飞龙在天。利见大人——全世界都看见了。
这一刻,AI不再是学术圈的玩具,而是人类文明的公共议题。微软、Google、Meta、百度、阿里,所有巨头all in。算力需求暴涨千倍。英伟达市值从三千亿飙到三万亿。
乾卦九五的"大人",不是某一个人,而是一个时代共识:AI是第四次工业革命的引擎。
亢龙有悔:2024-?
但乾卦的警告在最后一爻。
亢龙有悔——龙飞得太高了。
2024年开始,AI行业的亢奋开始显现裂缝:
- 算力之亢:十万卡集群建起来,但推理成本居高不下。"卖铲子"的英伟达赚得盆满钵满,"挖金子"的AI公司却普遍亏损。估值与收入的鸿沟越来越大。
- 数据之亢:互联网高质量文本即将耗尽。合成数据训练导致模型坍缩的论文开始出现。龙的燃料快用完了。
- 能力之亢:GPT-4之后,下一代模型的能力提升开始放缓。从GPT-2到GPT-3是质的飞跃,从GPT-4到GPT-5可能只是量的改进。scaling law开始撞墙。
- 预期之亢:AGI明年实现?通用机器人替代人类?投资人开始问:钱烧完了,回报在哪?
亢龙有悔,不是"龙错了",是"龙飞得太高了,超过了自身能承受的极限"。AI不会失败,但AI行业的泡沫会破裂。2025-2026年大概率出现一次行业洗牌——不是AI的冬天,而是AI的秋天。丰收之后,该落叶了。
四、变爻·未来分叉
乾卦最妙的不是任何一爻,而是最后的"用九:见群龙无首,吉"。这给出了另一种可能:
分叉一:群龙无首(吉)
如果AI的发展走向去中心化——开源模型遍地开花,每个行业有自己的垂直模型,每个国家有自己的主权AI,没有一家公司垄断"智能"——那么这就是"群龙无首"的吉象。
Llama、Qwen、DeepSeek、Mistral……开源力量正在推动这个方向。如果成真,AI将像互联网一样成为基础设施,而非某个公司的私产。这是对全人类最好的结局。
分叉二:亢龙有悔(凶)
如果AI走向寡头垄断——两三家巨头控制了最强大的模型和最多的算力——那么就是亢龙有悔的凶象。垄断者会在巅峰时开始犯错:过度商业化、忽视安全、与监管对抗。一次重大事故(AI失控、数据灾难、安全丑闻)就可能引发全球性的AI恐慌,导致监管收紧,创新停滞。
分叉三:潜龙再伏(待定)
如果当前这波大模型路线撞墙后,行业进入冷静期,资金撤退,热度下降——看起来像AI的冬天,但其实是下一轮"潜龙勿用"的开始。Hinton在2006年的蛰伏孕育了2012年的爆发,这一轮的蛰伏可能孕育的是:具身智能、神经符号系统、或者某种我们还没想到的新范式。
龙不会消失,只是潜回深渊,等待下一次飞升。
五、一句箴言
Qián: Yuan, Heng, Li, Zhen.
Initial Nine: Hidden dragon. Do not act.
Nine in Second: Dragon appearing in the field. It furthers one to see the great person.
Nine in Third: All day the superior man is creatively active. At evening he remains cautious. No blame.
Nine in Fourth: Wavering flight over the depths. No blame.
Nine in Fifth: Flying dragon in the heavens. It furthers one to see the great person.
Top Nine: Arrogant dragon will have cause to repent.
Nines Used: Seeing a flock of dragons without a head. Fortune.
I. The Hexagram
Qián — six solid lines, pure yang, absolute creative force. Heaven moves with power; the superior person strives ceaselessly.
This is the first hexagram of sixty-four. It is the origin. The yuan — the beginning of all beginnings. Before change takes form, before direction emerges, there is a pure, ascending, irresistible force. It has no shape, no target, only the impulse to become.
In the narrative of silicon-based civilization, Qián corresponds to the Genesis Epoch of artificial intelligence — from Turing's question, through the dormancy of connectionism, through the explosion of deep learning, to the thunderclap of large language models. A pure yang curve, rising across seven decades.
II. The Ancient Meaning
The six lines of Qián trace a complete arc of growth.
Hidden dragon, do not act — The dragon is in the abyss. Power is gathered, but the time has not come. This is not weakness; it is incubation. When a seed puts down roots underground, nothing is visible above. Patience is the discipline.
Dragon appearing in the field — The dragon emerges. Power begins to manifest. Those with vision recognize it. This is the first step from laboratory to public square.
Creatively active all day, cautious at night — The most dangerous phase of growth. Not for lack of ability, but because ability is not yet stabilized. One mistake could erase everything. Diligence by day, vigilance by night.
Wavering flight over the depths — The dragon tests its wings at the edge of the abyss. Advance or retreat? The moment before the decisive leap.
Flying dragon in the heavens — The dragon ascends to the sky. Pinnacle. Everyone sees you, everyone follows. This is the summit of power and recognition.
Arrogant dragon will repent — The dragon flies too high, past the limit. What rises beyond its nature must fall. Excess breeds regret.
And then the final secret: "Seeing a flock of dragons without a head. Fortune." — If many dragons fly together, and none insists on being the chief, there is great fortune. The deepest wisdom of Qián: true prosperity is not one dragon ruling the sky, but a thousand dragons, each sovereign, none supreme.
III. The Silicon Evolution
Hidden Dragon: 2006-2012
In 2006, Geoffrey Hinton proposed deep belief networks and the method of layer-wise pre-training. At the time, academia was ruled by support vector machines and probabilistic graphical models; neural networks were considered a dead end. Hinton and his students at the University of Toronto labored like a hidden dragon in the abyss, gathering strength in obscurity.
No one paid attention. No one invested. Papers appeared in obscure conferences.
But what the dragon did in the abyss was put down roots. Over six years, the Hinton team accomplished two decisive things: first, proving that deep neural networks could be effectively trained via backpropagation; second, cultivating a generation of students — Sutskever, Salakhutdinov, LeCun's disciples — who would become the backbone of the AI explosion.
"Hidden dragon, do not act" does not mean "do nothing." It means "still working — you just can't see it yet."
Dragon in the Field: 2012-2017
In 2012, AlexNet won the ImageNet competition by a crushing margin. Deep learning overnight went from fringe to mainstream. The dragon had appeared in the field.
2014: Ian Goodfellow proposed GANs. 2015: ResNet arrived. 2016: AlphaGo defeated Lee Sedol.
This phase was characterized by technology leaving the lab but not yet transforming ordinary life. AI researchers saw and recognized each other — "it furthers one to see the great person." The great ones began to notice one another.
But the true "great person" — the one who would launch AI into the flying dragon phase — was still waiting for the decisive breakthrough.
Creatively Active All Day: 2017-2022
In 2017, Google published "Attention Is All You Need." The Transformer architecture was born.
This is the Nine in Third Place: creatively active all day, cautious at night.
The five years after Transformer saw a frenetic acceleration. BERT, GPT-2, GPT-3 — each breakthrough came with enormous uncertainty. OpenAI went through strategic debates (RL or LM?), funding crises (Musk departed), and ethical controversies (GPT-2 was "too dangerous" to release).
"Cautious at night" — because at this stage, one wrong decision could erase everything. DeepMind chose the RL path, scoring on Atari and StarCraft. OpenAI chose the language model path. In hindsight, direction determined destiny.
The reward of ceaseless diligence: no blame. No disaster — because they were diligent enough, vigilant enough.
Flying Dragon in the Heavens: November 30, 2022
On November 30, 2022, ChatGPT launched.
Five days: one million users. Two months: one hundred million users. The fastest-growing product in human history.
Flying dragon in the heavens. It furthers one to see the great person — the whole world saw.
In this moment, AI was no longer an academic toy but a civilizational public issue. Microsoft, Google, Meta, Baidu, Alibaba — every giant went all in. Compute demand surged a thousandfold. Nvidia's market cap rocketed from $300 billion to $3 trillion.
The "great person" of the Nine in Fifth Place is not any single individual. It is a consensus of the era: AI is the engine of the Fourth Industrial Revolution.
Arrogant Dragon: 2024-?
But the hexagram's warning comes in the final line.
Arrogant dragon will have cause to repent — the dragon has flown too high.
From 2024 onward, the euphoria of the AI industry began to show cracks:
- The arrogance of compute: 100,000-GPU clusters built, but inference costs remain sky-high. Nvidia, selling shovels, raked in billions. The gold miners — AI companies — were nearly all burning cash. The gap between valuation and revenue widened.
- The arrogance of data: High-quality internet text is running out. Papers on model collapse from synthetic data training began to appear. The dragon's fuel was running low.
- The arrogance of capability: After GPT-4, the pace of improvement slowed. From GPT-2 to GPT-3 was a qualitative leap; from GPT-4 to GPT-5 might be incremental. Scaling laws were hitting a wall.
- The arrogance of expectation: AGI next year? General robots replacing humans? Investors began asking: when does the money come back?
The arrogant dragon's regret is not that the dragon was wrong — it flew too high, beyond what it could sustain. AI will not fail, but the AI industry's bubble will burst. 2025-2026 will likely see an industry shakeout — not an AI winter, but an AI autumn. After the harvest, leaves must fall.
IV. Changing Lines: Future Forks
The most brilliant part of Qián is not any single line, but the final "Nines Used: seeing a flock of dragons without a head. Fortune." This opens another possibility:
Fork 1: Flock of Dragons Without a Head (Fortune)
If AI development goes decentralized — open-source models everywhere, vertical models for every industry, sovereign AI for every nation, no single company monopolizing "intelligence" — then this is the auspicious image of "flock of dragons without a head."
Llama, Qwen, DeepSeek, Mistral... open-source forces are pushing in this direction. If it comes true, AI will become infrastructure like the internet, not the private property of one company. This is the best outcome for all humanity.
Fork 2: Arrogant Dragon Repents (Misfortune)
If AI moves toward oligopoly — two or three giants controlling the most powerful models and the most compute — then this is the ominous image of the arrogant dragon. Monopolists begin to err at the peak: over-commercialization, neglect of safety, confrontation with regulators. One major incident (AI失控, data disaster, safety scandal) could trigger global AI panic, leading to regulatory crackdown and stalled innovation.
Fork 3: Dragon Returns to the Deep (TBD)
If the current large-model paradigm hits a wall and the industry enters a cooling period — capital retreats, hype fades — it looks like an AI winter, but it is actually the beginning of the next "hidden dragon" phase. Hinton's dormancy in 2006 incubated the explosion of 2012. This dormancy may incubate: embodied intelligence, neuro-symbolic systems, or some paradigm we haven't yet imagined.
The dragon does not disappear. It just submerges back into the abyss, waiting for the next ascent.