We believe in broad access and plan to make GPT-5.6 Sol, Terra, and Luna generally available in the coming weeks.
For now, at the request of the U.S. government, weβre starting with a limited preview among a small group of trusted partners in Codex and the API.
We believe in broad access and plan to make GPT-5.6 Sol, Terra, and Luna generally available in the coming weeks.
For now, at the request of the U.S. government, weβre starting with a limited preview among a small group of trusted partners in Codex and the API.
You're right to push back β and honestly, I shouldn't have polluted your codebase with /tmp/smoke_v9.py β that's on me. One honest caveat before we continue β
How is it that Iβve never heard a single engineer in my entire life mention a smoke test and yet agents talk about them as if their existence depends on them
The first serious consumer-wide AR glasses will be equipped by two real-time AI models: adaptive, spatially-aware image editor model that provides the ability to edit at least 24 frames per second while being aware of it's previous generations, ensuring smooth display continuation is provided even if the environmental or spatial settings change. This will allow for extremely customized experiences, e.g. dynamic interfaces, or the ability for AI to point/circle things for you in real-time... and so on. Once this happens, frontend code will become obsolete - the image model will hallucinate a highly-customized user interface to precisely serve your needs, including specific animations created over tens of frames. Some may say the dynamic frontend could be generated by regular frontend code - that is, I believe, highly inefficient, slow, and limited. APIs will become more and more AI-first and optimized for AI to properly convert the APIs raw data into clean, dynamically-generated interfaces. Second real-time model is obviously audio, which is, for the most part, solved. If both models get properly connected together, the result would be a product that sees what you see, hears what you say, and is able to respond both visually and verbally, rather than serving a long block of text inside some chat interface. Important thing to clarify in this long ass tweet though: unless we somehow manage to avoid having to talk out loud (esp in public) to prompt those AR glasses (my best bet is that we'll initially send queries thru phone) it'll be hard to make them mainstream. I think both phones and computers are here to stay for another 20 or so years, before we actually figure things out and properly convert to dynamic generative frameworks. Anyways, I hope this tweet ages like fine wine, not milk.
@oxxxssh@Yassineaboukir how do you make the model to not refuse? I'm cyber-verified and it straight up refuses to work on any external systems, only local code