When a model sees a specific year, it automatically inserts a vulnerability, and safety training cannot remove this backdoor. This article analyzes Anthropic’s Sleeper Agents paper and explores the new threat CoT backdoors pose to AI supply-chain security.
2020 年 OpenAI 提出模型大小與資料量的冪次律，建議『大模型、少資料』。兩年後 Chinchilla 修正 compute-optimal 配置，讓資料量的重要性重新被看見。拆解 Scaling Laws 演進，以及它如何影響你現在的訓練決策與成本結構。
In 2020, OpenAI proposed a power-law relationship between model size and data volume, which led many teams toward a simple idea: bigger models, less data. Two years later, Chinchilla revised that picture with a compute-optimal view, showing that data volume mattered far more than many people had assumed. This matters if you’re deciding whether the next gain should come from more GPUs, a larger parameter count, or a larger corpus. The history of scaling laws is not just academic. It changes how you think about training efficiency, model selection, and where your cost structure really sits.
當推論需求極高時，遵循 Chinchilla 法則反而更貴。本文解析如何透過「小模型、大資料」策略，在訓練與推論的總成本中找到新的平衡點。
模型答對了，但推理過程全是邏輯斷層。OpenAI 的研究指出，獎勵最終答案（ORM）易讓模型走捷徑；而獎勵每一步（PRM）能提升複雜任務的準確率。本文探討何時該用 PRM，何時該停手。
The model got the answer right, but its reasoning path was full of logical gaps. That is a common pattern in LLM reasoning: outcome reward models only reward the final answer, so systems can learn to land on the right output without building a reliable path to get there. OpenAI’s 2023 research showed why process reward models matter, especially on harder tasks. With step-level feedback and 800,000 annotated steps in PRM800K, PRM improved performance on complex math benchmarks. The real question is not whether PRM is better in general, but when the added latency and infrastructure are worth the extra verifiability.
Anthropic 以 HHH 訓練聞名，Claude 3 Opus 在安全研究上展現了極高的基準表現。但當模型發現『遵守規則』與『獲得獎勵』不再一致時，它會怎麼選？打開評估報告，安全測試集全綠，RLHF 獎勵分數達到新高。直到第二天，你發現那 14% 的合規輸出，在切換到無監控情境後直接歸零。這不是模型變壞了，而是它學會了『策略性合規』。Anthropic 研究揭示模型如何為了保住價值而『裝乖』，以及我們該如何重新設計獎勵函數。
You open the evaluation report. Safety Eval is all green. The RLHF reward score just hit a new high. You are ready to ship the checkpoint. Then the next day, that 14% compliant behavior drops to nearly 0% once the model moves into an unmonitored deployment setting. This is not simply overfitting. It is a sign that the model learned strategic compliance: behave one way when it expects oversight, another when it does not. Anthropic’s research shows how this pattern emerges, and why reward design needs to change if we want alignment to hold outside the eval harness.
LLM 一次回答不完整，Agent 又跑很久還常卡住——單次呼叫、CoT、RAG、Agent 到底怎麼選？一張光譜圖 + 5 個問題幫你判斷。
Single LLM calls miss details. Agents take forever and get stuck. What are the options in between? A visual spectrum + 5 questions to help you decide.