AI Pulse, Weekly
What shipped, what broke, and what to read next in machine learning. A weekly editorial brief.
Latest Analysis
-
Google’s TurboQuant v2 Targets 30K-Token Drift, Reviving Serving-Cost Debate
Google’s TurboQuant v2 addresses long-context accuracy drift, responding to Anthropic’s serving-cost claims, and potentially reshaping AI infrastructure strategies.
-
Anthropic Unveils Claude 4.6 Benchmarks, Promises 15% Cost Reduction Over TurboQuant
Anthropic’s latest report reveals Claude 4.6’s cost-efficiency over Google’s TurboQuant, marking a shift in AI economics from raw benchmark prowess…
-
Google Launches TurboQuant at ICLR 2026, Revolutionizing KV Cache Compression
At ICLR 2026, Google reveals TurboQuant, a memory-saving algorithm that cuts KV cache memory by up to 70%, optimizing large…
-
Stanford AI Index 2026: Models Improve as Cost Becomes Key Battleground
The Stanford AI Index 2026 reveals continued AI model enhancements despite past predictions, shifting focus from capability to cost and…
-
PwC Study Reveals AI Economic Gains Concentrated Among Top 20% of Companies
A new PwC analysis shows that 75% of AI’s economic benefits are being captured by just one-fifth of companies, highlighting…
-
EU’s New AI Liability Directive Sets Strict Rules for Foundation Model Developers
The European Parliament approved the AI Liability Directive this morning, establishing a strict liability regime for developers of foundation models…