#DSPy
3 articles tagged DSPy
Tagged articles
Articles tagged DSPy, newest first.
AI PerformanceAI Agent Tuning: When to Rewrite Prompts and When to Retrain Weights
Prompt tuning works with hundreds of rollouts; weights need thousands of examples. Optimization requires data, a built system and an evaluation metric together. GEPA raised aggregate benchmark scores 12.19 points over a GPT-4.1 mini baseline. Qwen3 on-policy distillation beat RL on AIME'24 with 1,800 GPU hours versus 17,920. Prompts fit API-only models and costly rollouts; weights fit small student models.
AI in PracticeTemperature, Attention and Context Rot: Model Mechanics Behind Agent Design
Sampling, attention and the KV cache set the limits agent builders design around. Temperature divides logits before softmax, so keep it low for math and code. OOLONG scores drop from near 90 percent at 8k tokens to 10 to 40 percent at 512k. Keep tools minimal and append new context at the end to keep the KV cache valid. Protect safety instructions from context compaction and test that they survive.
AI in PracticeAI Contract Review: Where Agents Fail on 40,000 Documents and How to Catch It
Polished-looking Claude Code output got two of five key contract figures wrong. An unlinked amendment hid $800,000 and a notice deadline came out a month late. The fix was a five-stage pipeline to structure contracts plus read-only MCP tools. Other traps: silent permission errors, double counting across agents, stale advice. Structure documents first, verify query logs in code, keep a human in the loop.
Get new articles by RSS or podcast
Add the RSS URL to your RSS reader or the podcast URL to your podcast app to get every new article as soon as it is published. The podcast reads out the headline and key points.
How to subscribe