Large Language Models
LLM JSON Output Is Messy: A Fault-Tolerant Layer Beats Switching Models
A practical parser layer can repair common JSON formatting errors more cheaply and reliably than replacing the model.
Back to home
Large Language Models
4 articles tagged with this topic
Prompt Engineering
Prompt Engineering Is Now a Hard Interview Bar — But Over-Engineering for Most
A leaked interview post makes Prompt Engineering a hard bar for AI engineers. We argue 3 of 5 paradigms are over-engineering for most enterprise daily
Agent
Long-Task Agents 'Lose Their Memory'—The Real Problem Is Unmanaged Context
Context ledgers assign each item a source, utility, expiry, risk, and eviction policy, keeping long-task agents reliable without ever-larger windows.
Prompt Engineering
Examples Beat Instructions for LLMs: Why Few-Shot Prompts Stabilize Output
Zero-shot vs. few-shot prompting determines AI output stability and dev costs. Using the right demonstration method drastically reduces ineffective AI