gabiteodoru·Oct 13, 2025Don’t Force Your LLM to Write Terse Code: An Argument from Information Theory for q/kdb+ DevelopersUpdate [October 20, 2025]: I’ve discovered a methodological error in the perplexity measurements below. I used an Instruct-tuned model…
gabiteodoru·Oct 6, 20250.6% Failure Rate: Solving LLM Code Generation for q/kdb+KX demonstrated that even the best LLMs struggle with q/kdb+ programming tasks, with failure rates of 57% on first attempt and 26% using…
gabiteodoru·Sep 10, 2025Getting LLMs to 100% Success on q/kdb+ HumanEval, and Why It Should Be the Baseline (Preliminary…Within the past month, KX has released two significant contributions to q/kdb+ AI tooling: first, they published a comprehensive evaluation…
gabiteodoru·Jul 31, 2025Building for LLMs: From Library Design to Cost OptimizationI’ve been working on my q/kdb+ coding assistant and the translation of q-sql into Python-like syntax recently. This got me thinking about…
gabiteodoru·Jul 17, 2025How I Taught LLMs to Write q/kdb+: Nested AI OrchestrationIn my previous post I looked at how q’s unusual evaluation order and operator precedence causes problems for LLMs trying to write q code…
gabiteodoru·Jul 7, 2025Why LLMs Can’t Write q/kdb+: Writing code Right-to-LeftIn my previous post I showed how LLMs struggle with coding in q/kdb+; having immediate feedback loops using the q-MCP server does help a…
gabiteodoru·Jun 24, 2025Making LLMs fix q/kdb+ syntax through iterationLLMs struggle with q/kdb+ as its syntax+idioms are very different from Python/JS/C++. While GPT-4 and Claude handle mainstream languages…
gabiteodoru·Apr 22, 2025Real-Time Pitch Shifting with the Phase Vocoder: A Complete Implementation GuideIntroduction