【分享】alex_prompter 的LLMs 观点

x.com @alex_prompter

Everyone assumes LLMs are the future of AI.
大家都认为大型语言模型是人工智能的未来。

The permanent foundation. The layer everything else gets built on.
永久的基础。其他一切都会在那层上层叠加。

I’m not so sure.
我不太确定。

The historical parallel that fits best isn’t the one most people want to hear.
最贴切的历史类比并不是大多数人想听的。

LLMs are Edison’s DC power grid:
→ Genuinely revolutionary
→ Commercially dominant
→ Solving real problems right now
→ But architecturally limited in ways that can’t be patched
LLMs 是爱迪生的直流电网：
→ 真正革命性
→ 商业主导地位
→ 正在解决真正的问题
→ 但架构上有无法修补的限制

Right domain. Wrong architecture. And the evidence is already here.
正确的领域。建筑风格错了。证据已经摆在这里。

Hallucination isn’t a bug. It’s the architecture.
幻觉不是虫子。是建筑风格。

Researchers have formally proven that LLMs cannot learn all computable functions and will therefore inevitably hallucinate when used as general problem solvers.
研究人员已正式证明，LLMs 无法学习所有可计算函数，因此在作为通用问题解决工具时不可避免地会出现幻觉。

That’s not a training data problem. That’s math.
这不是训练数据的问题。那是数学。

A separate paper demonstrated that hallucinations stem from the fundamental mathematical and logical structure of LLMs, making it impossible to eliminate them through architectural improvements, dataset enhancements, or fact-checking mechanisms.
另一篇论文证明，幻觉源于 LLMs 的基本数学和逻辑结构，使得通过架构改进、数据集增强或事实核查机制无法消除幻觉。

And here’s the part that really gets you:
There’s a direct link between hallucination and creativity in LLMs.
这里最让你感动的是：
幻觉与创造力之间存在直接联系。

It may be impossible to eliminate hallucination without impairing the model’s most crucial capabilities.
在不影响模型最关键能力的情况下，可能无法消除幻觉。

→ The thing that makes LLMs creative is the same thing that makes them lie
→ Fix one, you break the other
→ That’s not a tradeoff you engineer away. That’s a design constraint.
→ 让大型语言模型有创造力的原因，正是它们说谎的原因
→ 修好一个，你弄坏另一个
→ 这不是你靠工程来弥补的权衡。这是设计上的限制。

DC power had the exact same structural problem. It couldn’t transmit electricity over long distances.
直流电力也有完全相同的结构问题。它无法长距离传输电流。

Not because the engineering was bad. Because the physics made it impossible.
不是因为工程技术不好。因为物理条件让这不可能。

You needed AC. A fundamentally different approach.
你需要空调。一种根本不同的方法。

The “AC power” of AI is already being built. And it has names.
人工智能的“交流功率”已经在构建中。而且它有名字。

This isn’t theoretical. People are already building the replacement architectures.
这不是理论上的。人们已经在构建替代架构。

Yann LeCun left Meta and raised $1 billion to prove LLMs are a dead end.
Yann LeCun 离开 Meta，筹集了 10 亿美元，证明大型语言模型是死胡同。

AMI Labs raised $1.03 billion in seed funding at a $3.5 billion valuation in March 2026, making it the largest seed round in European history.
AMI Labs 于 2026 年 3 月以 35 亿美元的估值筹集了 10.3 亿美元的种子资金，成为欧洲历史上最大的种子轮融资。

His thesis is simple: LLMs predict the next word. That’s not intelligence. That’s autocomplete at scale.
他的论点很简单：大型语言模型预测下一个词。那不是智力。这就是大规模的自动补全。

His core technology, JEPA, operates in latent space, learning abstract representations of reality rather than surface patterns.
他的核心技术 JEPA 运行在潜在空间，学习现实的抽象表征而非表面模式。

LeCun used a vivid analogy: using an LLM to understand the real world is like teaching someone to drive by just talking.
LeCun 用了一个生动的比喻：用 LLM 理解现实世界，就像仅凭说话教别人开车一样。

A Turing Award winner didn’t just write a paper about it. He quit his job and bet a billion dollars on it.
图灵奖得主不仅仅是写了一篇论文。他辞职并押上了十亿美元。

Mamba is proving transformers aren’t the only game in town.
Mamba 正在证明变形金刚并不是唯一的选择。

Mamba achieves 5x higher throughput than Transformers with linear scaling in sequence length.
Mamba 的吞吐量是变压器（Transformers）的 5 倍，且序列长度实现线性缩放。

Thanks to intensive research in 2023-2025, non-transformer architectures have reached parity with Transformers on key language benchmarks, and in some cases surpassed them.
得益于2023-2025年的深入研究，非变换器架构在关键语言基准测试上已与变换金器保持平衡，甚至在某些情况下超越了变换金器。

Hybrid architectures are already shipping.
混合架构已经开始发布。

By 2026, models built on hybrid transformer-SSM architectures can ingest hundreds of pages of text at once, far beyond vanilla GPT-3 or GPT-4.
到 2026 年，基于混合变换器-SSM 架构的模型可以一次性吞入数百页文本，远超普通 GPT-3 或 GPT-4。

The alternatives aren’t coming. They’re here.
替代方案不会出现。他们来了。

Meanwhile, look at what the industry is building to keep LLMs functional:
与此同时，看看行业正在打造什么来保持 LLM 的正常运行：

→ Agents (because the model can’t verify its own outputs)
→ Tool use (because the model can’t interact with the real world)
→ Reasoning chains (because the model can’t reason natively)
→ RAG (because the model can’t reliably recall facts)
→ 代理（因为模型无法验证自身输出）
→ 工具使用（因为模型无法与现实世界互动）
→ 推理链（因为模型本身无法推理）
→ RAG（因为模型无法可靠地回忆事实）

These aren’t features. These are workarounds.
这些不是特色。这些都是权宜之计。

When you need that many patches, you’re running longer DC power lines and wondering why the voltage keeps dropping.
当你需要这么多接线时，你会拉更长的直流电源线，却在想为什么电压一直下降。

Now the part everyone actually needs: which skills survive the transition?
现在说说大家真正需要的部分：哪些技能能在转变中存活下来？

When DC shifted to AC, some electrical engineers thrived and some went extinct.
当直流电转向交流电时，一些电气工程师兴旺起来，有些则消失了。

The ones who thrived understood circuits, load management, and power distribution at a fundamental level. Those principles worked on any architecture.
那些成功的人对电路、负载管理和电力分配有着根本的理解。这些原则适用于任何架构。

The ones who didn’t? They only knew DC-specific wiring.
那些没做到的？他们只懂直流专用的布线。

The same split is coming. And it’s coming faster than people think.
同样的分裂也即将到来。而且它来得比人们想象的还要快。

Here are the skills that transfer no matter what replaces transformers:
以下是无论更换变压器如何都能转移的技能：

→ Systems thinking for AI workflows. Breaking complex tasks into steps an AI can execute. This works whether the AI is a transformer, an SSM, JEPA, or something we haven’t built yet. Architectures change. The need for structured task decomposition doesn’t.
→ AI 工作流程中的系统思维。将复杂任务拆解成 AI 能够执行的步骤。无论 AI 是变压器、SSM、JEPA 还是我们还没建成的东西，这种方法都适用。架构会变化。结构化任务分解的需求则不然。

→ Evaluation and verification. Knowing if AI output is right. LLMs have a “Self-Correction Blind Spot” where they can recognize errors but lack the reasoning pathways to correct them.  Whatever comes next will still need humans who can evaluate quality. This skill gets MORE valuable, not less.
→ 评估与核实。知道 AI 输出是否正确。大型语言模型有一个“自我纠正盲点”，它们能识别错误，但缺乏纠正它们的推理路径。无论接下来发生什么，仍然需要能够评估质量的人。这项技能的价值会提升，而不是降低。

→ Data literacy. Understanding what data an AI needs, how to structure it, what’s clean vs. noisy. Every AI architecture runs on data. Past, present, future. The people who understand data will always have leverage.
→ 数据素养。理解 AI 需要哪些数据，如何构建数据，什么是干净的，哪些是噪声的。每一个 AI 架构都基于数据运行。过去，现在，未来。懂数据的人永远拥有影响力。

→ AI-augmented workflow design. Not “how to write a good prompt” but “how to redesign a business process so AI handles the right parts and humans handle the right parts.” This is architecture-agnostic. It transfers to anything.
→人工智能增强工作流程设计。不是“如何写出好提示词”，而是“如何重新设计业务流程，让 AI 处理正确的部分，人类负责正确的部分”。这与架构无关。它可以转移到任何东西上。

→ Domain expertise + AI fluency. The most powerful combination is stacking AI fluency on top of deep domain expertise.  A lawyer who understands AI beats a prompt engineer who doesn’t understand law. Every time. Regardless of what model they’re using.
→ 领域专长+人工智能流利度。最强大的组合是将 AI 的流畅度与深厚的领域专业知识叠加。懂 AI 的律师胜过不懂法律的提示工程师。每次都是这样。无论他们用什么型号。

→ Clear problem definition. Prompt engineering is just one implementation of a deeper skill: translating human intent into machine-executable instructions. Whether that instruction is a prompt, an API call, a config file, or something that doesn’t exist yet, the ability to define what you want is permanent.
→ 明确的问题定义。提示工程只是更深层技能的一种实现：将人类意图转化为机器可执行的指令。无论这个指令是提示符、API 调用、配置文件，还是尚未出现的东西，定义你想要什么的能力都是永久的。

And here’s what DOESN’T transfer:
→ Memorizing specific model behaviors (“Claude does X, GPT does Y”)
→ Platform-specific tricks that only work on one tool
→ Building your identity around a single product name
→ “Prompt engineer” as a job title instead of a thinking skill
The difference is simple:
→ Transferable skills = understanding WHY something works
→ Non-transferable skills = memorizing HOW a specific tool works
而以下内容是不会转移的：
→ 记忆特定模型行为（“Claude 做 X，GPT 做 Y”）
→ 仅适用于单一工具的平台专属技巧
→ 围绕单一产品名称构建你的身份
→把“提示工程师”作为一个职位名称，而不是思考技能
区别很简单：
→ 可转移技能 = 理解某物为何有效
→ 不可转移技能 = 记忆特定工具的工作原理

WHY survives paradigm shifts. HOW doesn’t.
WHY 能经受范式转变影响。但 HOW 不行。

The bottom line
The principle behind LLMs is permanent. The architecture probably isn’t.
结论
LLM 背后的原则是永久性的。建筑结构可能不是。

That’s not bearish on AI. That’s the most bullish take possible. It means the best is still ahead of us.
这并不是对人工智能的悲观。这是最乐观的观点。这意味着最好的还在前方。

Use LLMs hard right now. Build with them. Ship on them.
现在就用大语言模型。和他们一起建设。发货。

But build your skills around the PRINCIPLES, not the PRODUCTS:
→ Learn systems thinking, not just prompting
→ Learn evaluation, not just generation
→ Learn data literacy, not just tool literacy
→ Learn workflow design, not just model tricks
→ Stack domain expertise on top of AI fluency
但要围绕原则而非产品来培养技能：
→ 学习系统思维，而不仅仅是提示
→ 学习评估，而不仅仅是生成
→ 学习数据素养，而不仅仅是工具素养
→ 学习工作流程设计，而不仅仅是模型技巧
→ 叠加领域专业知识，同时具备人工智能流利度

The people who do this will thrive in the transformer era AND whatever comes after it.
那些做这些的人会在变形金刚时代以及之后的时代中茁壮成长。

Edison built a working power grid that lit up Manhattan. It was real, valuable, and changed the world.
爱迪生建造了一个能照亮曼哈顿的电网。它是真实的、有价值的，改变了世界。

AC still replaced it.
交流電 (Alternating Current)还是取代了它。

LLMs Will Always Hallucinate, and We Need to Live With This

个人观点，LLMs属于力大飞砖，后面一定会有更奇妙的创新！