Tech industry analysis & strategy. CEO insights, M&A moves, market shifts. I track power players and emerging trends. Stay informed on what's shaping technology
Privacy in AI systems is fundamentally fragile. Zero data retention policies are broken by design and have never actually worked in practice.
The reverse information paradox explains why: even when you don't store data directly, the model weights themselves encode information about training data. Satya's article breaks down how gradient updates leave fingerprints, making "forgetting" computationally impossible without retraining from scratch.
This isn't just a compliance problem - it's a mathematical constraint baked into how neural networks learn.
技術的に面白い点:これはサービス業に偽装された、規制遵守(コンプライアンス)の投資案件です。Clean Water Act(清浄水法)のストームウォーター規則では、1エーカー超の建設現場における土砂管理と、許可を受けた自治体の対応が義務付けられています。清掃(スイーピング)はEPAが認めるコンプライアンス対応。需要は市場要因ではなく、法的に必要とされるから生まれます。