When people talk about LLM failures, the default word is usually "hallucination." But after building...
Most LLM production bugs stem from contract failures—skipped tool calls, schema violations, missing evidence—rather than hallucinations. This reframes debugging from 'hallucination' to specific workflow breaches, enabling targeted fixes and reproducible regression tests.
When people talk about LLM failures, the default word is usually "hallucination."
But after building and testing LLM apps, I think many production bugs are better described as contract failures.
A hallucination is when the model makes something up. That matters, but it is not the only failure mode.
The subtler bugs happen when the model had enough context, but violated the surrounding system contract.
Examples:
Every LLM hallucinates, and it is not a bug the next model release will fix. Next-token prediction...
Hallucination isn't a bug in LLMs — it's the mechanism. The real problem is misallocating non-determinism, and why "SKILLS.md is…
I do a lot of research. Legal documents, technical specs, academic papers, regulatory filings....
Most LLM failures are easy to describe and surprisingly hard to reproduce. A user reports that the...
LLMs hallucinate structure. A production agent needs a validation layer between the raw model output and the code that acts on…
I had a model refactor part of a backtesting engine last month. Gave it function signatures, call...
