Weaviate Podcast: Forgetting Is the Key to Scaling Agent Discovery, With Berkeley's Shangyin Tan
CShorten30 · x · 2026-11-05
Weaviate Podcast #149 features UC Berkeley PhD student Shangyin Tan discussing "Scaling Discovery," centered on the claim that long-running optimizers stall when an agent remembers too much—and forgetting can get them exploring again.
The conversation starts from GEPA and its successor optimizeanything, which extends these ideas from prompt optimization to code, GPU kernels, and agent harnesses. A key insight: prompts and harnesses are really generalization tests, whereas a single kernel is pure optimization—illuminating how optimization algorithms extend from models to code.
Chapters cover what discovery is, forgetting in optimization, multi-agent AI, diversity in optimization, just-in-time databases, expensive evals in auto research, forgetting in AI, and the "mismanaged genius" hypothesis.
More from coding & agent
- Together Compiler uses agents to continuously optimize inference, up to 2.6x faster — simran_s_arora · 2026-11-05
- OpenAI upgrades Codex: marked-up images now pinpoint edits via API — OpenAIDevs · 2026-11-05
- Baseten Router cuts coding-task costs 72% with only 0.8pt pass-rate drop vs Opus 5.5 — baseten · 2026-11-05
- BrowserForge: open-source MCP server gives agents your real browser with human approval gates — mostafazog · 2026-11-05
- Free course teaches analytics engineers AI skills beyond Snowflake and dbt — Zachly · 2026-11-05
- NMSLIB author publicly slams new vector search preprint for downplaying prior work — srchvrs · 2026-11-05