openai.comarticleHow we built ChatGPT’s novel ad stack for user trust and relevance
OpenAI's ChatGPT ad stack uses a custom post-trained LLM, the orchestrator, that makes tool calls for retrieval, filtering, scoring and auction nomination across stateful ad sessions. Ad selection runs in a separate service from answer generation, so ads cannot influence responses or delay them. The post covers ANN retrieval, a model-generated preamble, delivery versus impression, logging, and model-based evaluation.
TL;DR Describes how OpenAI built ChatGPT ads around an LLM orchestrator that calls retrieval, ranking and auction tools, isolated from answer generation.
Organisations named here
- OpenAI
- ChatGPT Ads
- Karthik Gopalratnam
- Ritendra Datta
Technologies named here
- Codex
- Ads orchestrator
- ChatGPT
- Ads Manager
- Ads orchestrator tool set
- LLM
Techniques named here
- Stateful ads sessions
- LLM orchestrator with tool calls
- Separation of ads selection from answer generation
- Eligibility filtering
- Candidate scoring and auction
- Time-bounded delivery with separate preamble
- Model-based evaluation and self-improvement loop
- Granular outcome logging
Related in agents
