The discussion centers on OpenAI's B2B platform, which originated as its first product before ChatGPT. The platform enables a wide range of companies, from startups to Fortune 500s and government agencies, to build on top of OpenAI's models, driving AI adoption across various sectors.
OpenAI is undertaking complex, mission-critical deployments that go beyond standard API calls. Examples include a custom, on-premise installation of models on the 'Venato' supercomputer for Los Alamos National Labs and deep collaboration with Amgen to augment drug development.
The launch of GPT-5 is presented as a major breakthrough in reasoning and performance, with claims of near-zero hallucination and the ability to solve complex problems in a single attempt. However, this advanced capability comes with trade-offs, including significant latency and the need for users to re-engineer prompts designed for less capable models.
The conversation details the shift from 'stitched' multimodal systems (e.g., separate speech-to-text, LLM, and text-to-speech models) to integrated, end-to-end solutions like the real-time voice API. This new approach reduces latency and preserves crucial conversational signals like tone and pauses, creating a more natural interaction.
Keep pulling the thread on Sherwin Wu & Olivier Godement.