Ex-Bing chief Mikhail Parakhin: LLMs inherit mediocre code taste from pretraining
MParakhin · x · 2026-10-05
Mikhail Parakhin argues LLMs are biased by pretraining on mediocre code; post-training helps but the flaws remain visible. His pet peeve: unnecessary heap allocation (including std::optional/Option), which he cleans up by running dedicated critique-loop iterations each time.
More from coding & agent
- Skills vs MCP vs RAG vs Memory: a 4-part framework for agent knowledge — MaryamMiradi · 2026-10-05
- GitHub Copilot CLI v1.0.92-4 adds config subcommands and a batch of stability fixes — copilot-cli-release-app[bot] · 2026-10-05
- Inner local agent injects context into edge agents via attestation — natesiggard · 2026-10-05
- Multiplayer tank shooter built with Claude Code: Opus orchestrator runs parallel Sonnet agents in git worktrees — IamHuggos · 2026-10-05
- Blog: Agents Don't Need Memory Plugins, They Need Documentation — itsOmSarraf_ · 2026-10-05
- Open-source StreamCut embeds an MCP server: four half-remembered movie quotes in, four trimmed clips out — chulobou · 2026-10-05