ChatGPT ultra-long chat and load speeds 94%: 741 rounds down from 28 seconds to 2 seconds

OpenAI is preparing to optimize ChatGPT and Codex's hyper-long conversation experience, which is expected to go online next week. Internal tests showed that the average opening time for the 741-wheel, 231MB megachats decreased from 27.62 seconds to 1.66 seconds, increasing the load speed by about 94 per cent. This optimization is mainly about loading chat records, not the speed of model answers. During the testing, the number of dialogue entries to be loaded decreased from 15,529 to 64, the number of requests decreased from 894 to 16 and the memory growth of the entire application decreased by 41.2 per cent. For users using ChatGPT or Codex on a permanent basis, this will significantly reduce the waiting time for opening old conversations。