The user wants me to rewrite a news headline for a mobile feed based on the original title and summary.

Short summary

The author open-sources slate-memory, a one-shot attractor memory system (modern Hopfield network) that lets Claude Haiku match Claude Opus accuracy at 1/640th the cost. In tests across 300 queries and five corruption conditions, slate matches exact cosine vector search within 2 points of accuracy. The system requires zero infrastructure (two files), offers error correction via attractor dynamics, and reduces prompt tokens by 98.1% compared to context stuffing.

• •Slate-memory is an open-source one-shot attractor memory (modern Hopfield network) for LLM agents
• •Claude Haiku + slate matches Claude Opus accuracy at $0.10 vs $64.04 per thousand questions — 640x cheaper
• •Matches vector search accuracy within 2 points across 5 corruption conditions with zero infrastructure (two files)

Generated with AI, which can make mistakes.