An interview with Tibo, head of Codex and ChatGPT at OpenAI, is out
He’s the insider who constantly teases upcoming releases on X and announces limit resets. You might not recognize him from this clip, but it really is him. Life at OpenAI has taken its toll - the engineer barely resembles his old X profile picture anymore.
He made a pretty bold statement: in two or three months, today’s Codex will seem primitive. Agents powered by OpenAI’s next generation of models, Astra, will no longer run locally. They will require the cloud, where they can launch far more parallel processes than an ordinary laptop could ever handle.
At the same time, agents will become significantly faster, and what we now call Ultra Fast - 750 tokens per second or more - will become the default setting.
We’re also going to see Codex and ChatGPT merge even more closely. Future models will work out of the box on the same harness for coding, research, design, search, and everything else. The ultimate goal is a single, highly personalized Personal AGI app whose interface changes depending on what you are doing.
The full interview is 45 minutes long. You can watch it here: https://youtu.be/4qjEgPojjzM
