lilian-weng
Lilian Weng — author of Lil’Log (lilianweng.github.io), one of the field’s most-cited technical explainer blogs, and of this wiki’s LLM-powered-autonomous-agents survey (Jun 2023).
What this wiki knows from the source
- She wrote the canonical early synthesis of the LLM-agent concept — the LLM brain + Planning + Memory + Tool use anatomy (llm-agent) that most later agent vocabulary descends from.
- The post’s writing process is itself notable: she credits ChatGPT for help drafting the Memory section (“I’ve learned a lot about the human brain and data structure for fast MIPS in my conversations with ChatGPT”) — an early, explicit instance of using the tool to write about the tool.
- The post is written from clear proximity to OpenAI’s stack (it treats ChatGPT Plugins and OpenAI function calling as first-class examples of tool-use).
Background (not established by this source)
Widely reported but not asserted by the ingested post: Lilian Weng led safety / applied-research work at OpenAI for several years before departing in late 2024, and is associated with the founding of a new AI lab in 2025. Flagged as background; wants a primary source before the wiki states it as fact.
Why she’s worth a page
Lil’Log posts recur as reference points across this domain (prompt engineering, RL, diffusion, hallucination, agents). This is the first ingested into the wiki; more will plausibly follow, and the agent survey is load-bearing enough that a stable author node is worth keeping.
Ingested sources
- 2026-06-04-llm-powered-autonomous-agents — the canonical 2023 LLM-agent survey (LLM brain + Planning + Memory + Tool use).
- 2026-07-07-harness-engineering-self-improvement — Jul 2026 survey on harness engineering for recursive self-improvement; covers context engineering (ACE, MCE), workflow search (ADAS, AFlow), self-improving harnesses (STOP, Self-Harness, DGM), evolutionary search (AlphaEvolve), and joint weight optimization (SIA).
Cross-link
- llm-agent — the anatomy the 2023 survey introduced.
- 2026-04-27-the-second-half-of-ai — companion lineage: Shunyu Yao (ReAct author) writing two years later on where the recipe leaves the field.
- recursive-self-improvement, harness, context-engineering — concepts the Jul 2026 post significantly extends.