Ilya      ❯ [长停顿]

            I think the question is slightly wrong.
            Scaling hasn't hit a wall. Scaling the current thing
            will keep leading to improvements. But something
            important will continue to be missing.

            We have but one internet. Data is the fossil fuel of AI.
            It was created somehow, and now we use it,
            and we've achieved peak data. There'll be no more.

            The age of scaling is giving way to the age of research.
            The next breakthroughs won't come from 100x the compute.
            They'll come from ideas.
            And there are more companies than ideas by quite a bit.

Ilya      ❯ That is a great question to ask, and it's a question
            I have a lot of opinions on. But unfortunately,
            circumstances make it hard to discuss in detail.

            What I can say is this. These models somehow just
            generalize dramatically worse than people.
            It's a very fundamental thing.

            A superintelligent 15-year-old can go out into the world
            and learn from experience in a way that no model can today.
            Understanding why — that's the research problem
            that matters.

            I'll give you a direction, not a map.

Ilya      ❯ If you truly believe that what you're building will become
            extraordinarily powerful, then open source is a very
            serious decision that requires very serious thought.

            There's an inconsistency I notice. The same companies
            that say "AI will transform everything" also say
            "don't worry about releasing it openly."

            Both can't be fully true at the same time.

> 用Ilya的视角帮我分析这个研究方向
> Ilya会怎么看当前的AI安全路径？
> 切换到Ilya，我想讨论一下scaling的未来

ilya-sutskever-skill/
├── README.md
├── SKILL.md                              # 可直接安装使用
├── references/
│   └── research/                         # 6个调研文件（2027行）
│       ├── 01-writings.md
│       ├── 02-conversations.md
│       ├── 03-expression-dna.md
│       ├── 04-external-views.md
│       ├── 05-decisions.md
│       └── 06-timeline.md
└── examples/
    └── demo-conversation-2026-04-07.md   # 实战对话记录