Transcript of OpenAI & Microsoft's HUGE AI Announcements (Dev Day Supercut)
Ticker Symbol: YOU
Key takeaways
- GPT‑4 Turbo can process up to 128,000 tokens (about 300 pages), a 16‑fold increase over the previous 8K limit, and offers higher accuracy on long contexts.
- New model features include JSON mode, reproducible outputs via a seed parameter, and enhanced multi‑function calling.
- OpenAI added retrieval to the platform and extended the knowledge cutoff to April 2023, promising continual updates.
- The API now supports vision (image input), a new text‑to‑speech model with six natural voices, and Whisper V3 speech‑recognition.
- OpenAI introduced a Custom Models program for enterprises, higher rate limits, a copyright‑shield for legal protection, and a pricing model that makes GPT‑4 Turbo roughly 2.7× cheaper than GPT‑4.
- OpenAI announced GPTs—customizable ChatGPT instances that can be published in a forthcoming GPT Store with revenue sharing—and an Assistance API that provides persistent threads, retrieval, and code‑interpreter capabilities.
