the log · 159 entries
Log-scale notes on infrastructure.
Analysis on backup, cloud, open source, security and the companies shaping them. One entry a day, filtered through twenty-five years of running the systems involved.
all 159
ai-machine-learning 16
ai-engineering 78
infrastructure 119
machine-learning 3
programming 64
software-development 40
technology 113
technology-leadership 29
tutorials 1
2026.08.088 min read
Terafab: Tesla and SpaceX's $16.8 Billion Plan to Control Chip Production
Tesla and SpaceX plan to bring chip manufacturing, memory, packaging, and testing into one Texas complex. The case rests on faster development and reliable supply, but building a working…
2026.08.077 min read
AMD Plans to Acquire Taalas, Betting on AI Models Built Into Silicon
AMD's planned Taalas acquisition targets a costly part of AI inference: moving model weights from memory. Taalas builds those weights into silicon, trading flexibility for claimed gains in…
2026.08.067 min read
GPUBreach: How GPU Rowhammer Reached Host Root with IOMMU Enabled
University of Toronto researchers demonstrated a path from GDDR6 bit-flips to host root access. GPUBreach uses permitted DMA buffers and NVIDIA driver flaws to get past the protection IOMMU…
2026.08.058 min read
Infinity's 10-Hour Chip Software Build Tests Nvidia's CUDA Advantage
Infinity reportedly built CUDA-like software for an inference chip in about 10 hours. That could lower switching costs, but production testing and Nvidia's broader software stack remain…
2026.08.047 min read
Andy Pavlo Joins ClickHouse to Build a Database Research Lab
Andy Pavlo is leaving Carnegie Mellon to establish ClickHouse Labs. The research agenda covers databases for AI agents, validation of existing engineering, and new hardware and execution…
2026.08.037 min read
Go 1.27 Adds Generic Methods, Post-Quantum TLS, and Stricter JSON Handling
Go 1.27 adds generic methods, post-quantum cryptography, and opt-in json/v2 APIs. Goroutine leak profiling and faster small allocations also give production teams reasons to test the…
2026.08.028 min read
GCC Rejects Significant AI-Generated Code, While LLVM Takes a Disclosure-Based Approach
GCC's new policy rejects significant AI-generated code, including revised versions of model output. Its exceptions and its differences with LLVM show how copyright and review requirements…
2026.08.017 min read
AI Coding in 2026: Why Developers Can Feel Faster While Output Barely Changes
A randomized trial found developers took 19% longer with AI while believing they were 20% faster. Broader data shows modest throughput gains, with results depending on the task and workflow.
2026.07.317 min read
AI Engineers Want a Way to Slow Frontier Development. That Deserves Support.
A July 28 letter signed by 1,178 AI researchers, engineers, and executives asks for tools to coordinate a slowdown if needed. The proposal deserves support, alongside closer attention to…
Nothing on this page matches. Press Enter to search every entry.