OpenAI identified three codex-based extra consumption points

OpenAI finally found out exactly why Codex fell fast. The director, Tibo Sottiaux, said the team has discovered three additional consumptions: It is inefficient to repeatedly compress the context with a long session with a picture; it is too high for a part of the high-consumption scene; and it is even more efficient to automatically generate a session title than expected. Tibo previously stated that no overall anomaly had been detected and that some of the affected users were directed to sub2api and shared with the subscriber. It was only yesterday that he further admitted that some users had indeed experienced a deterioration in their CLA, which could lead to increased consumption. Now the investigation has finally located Codex's own specific problems. The aim of this "constriction of context" is to keep Codex compressing old content during a long mission, free of context. OpenAI also confirmed that Codex will automatically use operation to maintain a long mission. This process is now a waste of extra time when the images are numerous and successively compressed. Computer History is a recent go-live feature that brings the selected App on Mac and the operating records on the web page to ChatGPT and Codex. OpenAI will send the repairs on Sundays of American time, while making a complete Restet for all paid subscriptions to Codex. Tibo says Reset is expected to go online around 2 p.m. US West Coast time, corresponding to Beijing time on August 24th, around 5 a.m. The team has also found a new, unrelated optimisation programme, which it claims will significantly improve efficiency and will continue next week。