2026

Part three measures Qwen3.8 Flash Next story generation, sustained throughput, and patched long-prompt performance on an Apple M4 Max.

• llm
• ollama
• benchmarking
• performance

Part two compares local LLM input, output, latency, memory, and sustained performance on an Apple M4 Max, with dated Claude and GPT-5.6 Sol API speed references.

• llm
• ollama
• benchmarking
• performance

Bourbaki Notes now has per-post discussions with local accounts, proof-of-work abuse control, and a small admin interface.

• site news
• comments
• security

A local Ollama benchmark checks whether language models can identify the shrinking state variable in a recursive integer-square-root complexity analysis.

• llm
• algorithms
• complexity
• benchmarking

Part four extends LifetimeCheck through memory operations, structured control flow, scope cleanup, coroutine tasks, and returned-lambda safety.

• clang
• llvm
• mlir
• static-analysis
• cpp

Part two follows the LifetimeCheck pass architecture, its type categories, points-to sets, allocation classification, and operation dispatch.

• clang
• llvm
• mlir
• static-analysis
• cpp

Part one introduces ClangIR as an MLIR-based representation for C++ static analysis and uses LifetimeCheck to motivate the series.

• clang
• llvm
• mlir
• static-analysis
• cpp

Part three shows how ClangIR AST attributes, move detection, smart pointer handling, and moved-from state tracking detect use-after-move bugs.

• clang
• llvm
• mlir
• static-analysis
• cpp