Codex, OpenAI’s New Coding Agent, Wants to Be a World-Killer

Although artificial intelligence takes the world through the storm, it is still very bad in tasks that require high elasticity of flexibility, such as writing a computer code.

Earlier this year, Chatgpt Maker Openai has published a white paper that Amnesty International takes the task of its dull performance in SCRUM coding. Among other things, it was found that even the most advanced artificial intelligence models “are still unable to solve the majority of” coding tasks.

Later in an interview, the CEO of Openai Sam Altman said that these models “on the identity of being incredible in software engineering,” adding that “software engineering by the end of 2025 appears completely different from software engineering at the beginning of 2025”.

It was a bold prediction without supporting it – if anything, the gym AI like the gentle Altman pedals are worse in coding while increasing hallucinations with every new repetition.

Now we know what he was playing.

Early on Friday, Openai revealed a preview of Codex, stabbing the company in a specialized “agent” – a delicate industrial term that seems to change the definitions depending on the company that is trying to sell it to you.

“Codex is the software engineering agent based on the group of cores that can work in many tasks in parallel,” says the company’s research preview.

The new tool will apparently help software engineers by writing new features, correcting the current code, and answering questions about software instructions, among other tasks.

Unlike the Chatgpt everything in-A-Box model, which is directed towards the comprehensive market, Codex has been trained to “create a symbol that closely reflects the human style and public relations preferences.” This is a charitable way to say “stealing the symbol of others”-an artificial intelligence training tactic in the non-distant past, when Copilot from Microsoft helped go to the city in an open source and subscriber symbol on Gabbab.

Thanks in a large part to the technique, Openai, GitHub and Microsoft came out of this legal fight largely safely, giving Openai some comfortable legal shields if it chooses to go alone through its internal model trained on the GitHub icon.

In the Codex version, Openai claims that his coding agent is working entirely in the cloud, and it is cut off from the Internet, which means that he cannot search for the web for data like ChatGPT. Instead, “Openai limits the agent’s interaction only on the symbol explicitly presented via GitHub warehouses and pre -installed consequences formed by the user via the preparation program.”

However, the data used to train Codex had to come from somewhere, and to judge the rash of copyright claims that appear to be P.The artificial intelligence industry is delayedIt is a matter of time only before we discover where.

More on Openai: Chatgpt users are developing strange delusions