On August 29th, local time, 27th, according to Connect magazineOpenAI I'm developing a more proactive, sustainable job CodexI WANT THE FLAGSHIP AI AgentNo longer limited to waiting for the user to give the job。
Connect to check Cordex's recent code changes and it turns out that OpenAI has been working on the Cordex command line version in recent daysAdd Enduring ModeI don't know. The code change for the Codex command line tool is default public, and OpenAI's new functionality will usually be here first, and then extended to products such as Codex Desktop Application and ChatGPT Work。
The durable model is still at the test stage and has not yet become widely available or officially published. OpenAI spokesmen have also confirmed that the company is testing the functionality, but there are no plans to go online in the near future。
"OpenAI's culture is very bottom-up, and you're going to try a lot of different things in the open-source library, and it's kind of like our common testing ground."
Lasting mode appears in the Codex 'Debtivity' menu. User canDetermine how much calculator, Token and time you invest before a model answers a question.
At the same time, a durable model or one of the most resource-consuming settings available to OpenAI. Once enabled, Codex willKeep working until you're in hibernationI don't know. Other models at present, even if the task is not completed, usually cease to operate in minutes or hours。
OpenAI also designed a capability called “active” for a durable model. According to the system hint in the code, when a smart person answers a user's question, the task cannot be considered closed, but needsTake the initiative to organize follow-up for yourselfI don't know. Intelligent bodies can continue to perform these tasks across sessions and determine what to do next, taking into account past interactions with users and “know-your-user” knowledge。
The system even supports intelligenceSend the message if the user does not request itHowever, the requirement is to be used as little as possible。
OpenAI has also defined the limits of access for this working intelligence. A long-lasting model does not expand the original operating privileges of an intelligent body if it needs to modify the content outside the user's own systemUser approval is required first.
The code for the "active" function is not a specific part of the Codex terminal tool, but is located in the Codex shared core. According to the report, this suggests that OpenAI or plans to bring this capability to more products。
According to 1AI, OpenAI CEO Sam Altman has recently spoken publicly in the hope of eventually turning ChatGPT into oneAN AI SMART BODY THAT CAN INITIATE AND SUSTAINI don't know. OpenAI also hopes to increase the use of state-of-the-art models, as the actual use of these models is still only a small part of ChatGPT ' s large user base。
OpenAI has tried a number of active AI products before and has never really opened the market. For example, the Pulse, which was launched last year, produces morning briefings during the user ' s sleep and was stopped early this summer。