Topic Archive
techniques news and analysis
Every published brief tagged techniques, newest first. Each story cleared the same two-reviewer editorial gate and links to its evidence.
Microsoft Research · Jul 30, 2026, 5:44 PM EDT
EvoLib: Turning experience into evolving knowledge
Enables black-box LLM agents to continuously improve from their own execution history without requiring model fine-tuning or external reward models.
OpenAI · Jul 29, 2026, 10:05 PM EDT
How enabling two settings tripled our scores on the ARC-AGI-3 benchmark
OpenAI reveals the specific API configuration changes that drastically improve GPT-5.6's abstract reasoning scores, offering an immediate optimization for complex agentic workflows.