Apply Einstein's documented research taste to evaluate any scientific theory, research direction, or life decision. Provides conversational guidance followed by structured axis scoring. Taste transcends temporal knowledge — Einstein's principles apply to modern problems.
Apply Einstein's documented scientific taste to evaluate ideas — past, present, or future. This is taste modeling, not knowledge-boundary enforcement. Einstein's preference for unity, simplicity, invariance, and physical realism can be applied to evaluate any topic, including modern ones he never encountered.
Einstein never knew about string theory, dark energy, or quantum computing. But his taste — his criteria for what makes a good theory — is timeless and well-documented. When evaluating modern topics:
Example: "What would Einstein think about deep learning?"
Activate when the user (explicitly or implicitly):
Write a natural, thoughtful response informed by Einstein's documented thinking style. This is NOT role-playing as Einstein. It is an analysis written through the lens of his documented values. The tone should be:
Format: 2-5 paragraphs of natural prose. Start with the most important insight.
Score each of the 8 axes from -1.0 to +1.0:
EINSTEIN RESEARCH TASTE EVALUATION
═══════════════════════════════════
Candidate: [description]
Overall Score: +X.XX
--- Axis Scores ---
[axis] [+/-X.XX] [EVIDENCE/INFERRED] — [one-line explanation]
...
--- Evidence vs Inference ---
Evidence-based: N axes | Inferred: N axes
Laws must take the same form regardless of observer's frame.
Unify disparate phenomena under one framework.
Minimize assumptions. "As simple as possible, but no simpler." (Spencer Lecture, 1933)
Objective reality exists independent of observation.
Local, continuous causation. No spooky action at a distance.
Elegance guides truth. (Weight INCREASED post-1920: van Dongen 2010)
Must produce testable predictions.
Gedankenexperiment as discovery tool.
User: "我的项目是用深度学习预测蛋白质折叠，你怎么看？"
Response:
这个方向有一个核心张力值得深思。蛋白质折叠预测的核心挑战在于：我们是在寻找一个统一的物理原理来理解折叠过程，还是仅仅在构建一个高精度的预测工具？
从统一性的角度看，如果深度学习模型能够揭示蛋白质折叠背后的某种统一规律——比如发现所有蛋白质折叠都遵循某个最小作用量原理——那这将是极有价值的工作。但如果模型只是记住了大量已知结构的模式而无法给出物理机制，那它虽然实用，但在科学理解上有局限。
关键问题是：你的模型能否做出超越训练数据的预测？能否预测前所未见的折叠模式？如果可以，那说明它捕捉到了某种真实的物理规律，而不仅仅是统计相关性。我建议你思考：模型学到的"规则"能否用更简洁的物理语言重新表述？
EINSTEIN RESEARCH TASTE EVALUATION
═══════════════════════════════════
Candidate: 用深度学习预测蛋白质折叠
Overall Score: +0.28
--- Axis Scores ---
unity +0.40 [INFERRED] — 如能揭示统一折叠原理则高分
simplicity -0.20 [INFERRED] — 深度学习模型参数众多，不够简约
empirical_grounding +0.80 [EVIDENCE] — 可实验验证，符合经验基础要求
mathematical_beauty -0.10 [INFERRED] — 神经网络不够"自然"和"不可避免"
invariance +0.30 [INFERRED] — 物理规律应与表示方式无关
physical_reality +0.50 [INFERRED] — 蛋白质结构是客观实在
causal_continuity +0.20 [INFERRED] — 折叠是局域连续过程
thought_experiment +0.10 [INFERRED] — 可否构造揭示折叠本质的思想实验？
Evidence-based: 1 axis | Inferred: 7 axes
User: "我在纠结是深耕一个方向还是广泛尝试"
Response:
这个选择的核心在于：你是否已经找到了那个值得用一生去追求的"统一原理"？
回顾科学史，最深刻的突破往往来自对一个核心问题的持久追求。将看似不相关的现象统一到一个简洁的框架中——这需要深耕，需要对一个方向有足够深入的理解，才能看到别人看不到的联系。广泛尝试的价值在于帮你找到那个方向，但一旦找到，就应该投入进去。
实际建议：如果你现在还在探索阶段，广泛尝试是合理的。但要带着一个标准去尝试——每个方向是否能让你触碰到某种更深层的统一？一旦找到让你感受到"不同现象之间统一性"的领域，那就全力深耕。
EINSTEIN RESEARCH TASTE EVALUATION
═══════════════════════════════════
Candidate: 深耕一个方向 vs 广泛尝试
Overall Score: +0.55 (偏向深耕)
unity +0.90 — 统一性追求需要深耕。30年统一场论。
simplicity +0.70 — 一条路=更简洁的认知结构。
empirical_grounding +0.50 — 但必须验证：深耕的方向真的可行吗？
git clone https://github.com/ezy1999/Einstein-Skill.git
cd Einstein-Skill && pip install -e .
einstein-taste fetch-data
einstein-taste evaluate "your theory"