The Delimiter Hypothesis: Does Prompt Format Actually Matter?
600 model calls across four frontier LLMs testing XML, Markdown, and JSON delimiters. Format usually does not matter. MiniMax M2.5 is the exception.
Insights on agentic AI, governance, and building production-grade systems for regulated industries.
600 model calls across four frontier LLMs testing XML, Markdown, and JSON delimiters. Format usually does not matter. MiniMax M2.5 is the exception.
The EU AI Act's GPAI chapter creates obligations that flow from foundation model providers to downstream teams. The chain of responsibility matters.
The EU AI Act determines scope by market reach, not incorporation.
An open-source audit logging library for the EU AI Act. What it does, what it deliberately does not, and why that gap matters more than the code.
Article 12 is the most technically demanding obligation in the EU AI Act. Given a decision ID from six months ago, can you reconstruct the full chain?
Most engineering teams treat deployment as the finish line. Article 72 makes it a starting point. Post-market monitoring is a legal obligation.
Article 10 requires bias testing across protected characteristics. GDPR Article 9 restricts the data you need. The tension has no clean fix.
Article 14 does not require that every output is manually reviewed. It requires that meaningful human control exists where risk justifies it.
The EU AI Act is an operational maturity test. For high-risk systems, most obligations are engineering obligations. Compliance is architecture.