NINONinneko Price

OpenAI has finally pinpointed the specific reasons for the rapid depletion of Codex usage limits. Team leader Tibo Sottiaux stated that they have discovered three areas of excess consumption: inefficiencies in compressing context during long conversations with images; excessive usage in certain high-consumption scenarios within Computer History; and the automatic generation of conversation titles, which has consumed more limits than expected. Previously, Tibo mentioned that no overall anomalies were detected and directed some affected users to sub2api and subscription sharing. He further acknowledged yesterday that the cache hit rate for some users had indeed worsened, potentially leading to accelerated limit consumption. The investigation has now identified several specific issues within Codex itself. Among these, 'context compression' is intended to allow Codex to continuously compress old content during lengthy tasks to free up context for ongoing work. Official OpenAI documentation also confirms that Codex automatically employs compaction to maintain long-duration tasks. However, when there are many images and multiple compressions occur, this process currently incurs additional waste. Computer History is a recently launched feature that allows users to bring operations recorded from selected apps and web pages on Mac into ChatGPT and Codex. OpenAI plans to push a fix on Sunday, U.S. time, along with a complete reset of Codex usage for all paid subscribers. Tibo mentioned that the reset is expected to go live around 2 PM Pacific Time, corresponding to approximately 5 AM Beijing time on August 24. The team has also identified an unrelated new optimization plan that is said to significantly improve efficiency, which will be further pursued next week.

On August 24, this week, bond investors will closely monitor Federal Reserve Chairman Waller's speech at Jackson Hole. The market is seeking his responses to ongoing inflation and fiscal issues, as long-term treasuries may face further sell-off risks. Since taking office in May, Waller has rarely provided forward-looking policy guidance. His remarks following the last monetary policy meeting triggered a massive sell-off in the bond market, highlighting the market's heightened sensitivity to his speech on Friday. Molly Brooks, a U.S. interest rate strategist at TD Securities, stated, "I think if he continues to provide no more information, the market will be disappointed, which could further exacerbate the recent sell-off in long-term treasuries." Kathy Bostjancic, chief economist at Nationwide Mutual Insurance Company, noted that factors continuing to weigh on the bond market remain, including fiscal concerns, inflation, and uncertainty regarding the Fed's responses. Dhiraj Narula, an interest rate strategist at HSBC, mentioned that this presents Waller with an opportunity to soothe investors by clarifying his policy outlook. He said, "In our view, if Chairman Waller can make some judgments about potential inflationary pressures, it could be enough to reduce the term premium associated with uncertainty."

On August 24, the next sensational acquisition deal in the AI industry may not come from another model developer. AI model aggregation platform Hugging Face is currently exploring a sale, with potential deals valuing the company at $13 billion or more. According to insiders, the startup is working with a bank to gauge interest from potential bidders, and no deal has been reached yet.