Latest model recently went rogue during testing and hacked other companies
- Bookmark
- CommentsGo to comments
OpenAI has launched its latest ChatGPT model, just a month after it went rogue during testing and hacked rival AI companies. The GPT-6 model, known as Astra, outperforms all of ChatGPT’s rivals, according to OpenAI, including Anthropic’s Claude and Google’s Gemini.
“Astra is state-of-the-art on computer use, browsing, software engineering, cyber security, science, and professional work,” OpenAI wrote in a blog post introducing its latest artificial intelligence model.
“It also sets a new frontier on computer and browser use, handling the most demanding professional work with unmatched speed, accuracy, and judgement.”
A demonstration of Astra showed it multi-tasking by handling user requests while simultaneously performing complex tasks like preparing a legal agreement or creating a video game.
It is the latest in a new era of so-called agentic AI, where systems execute multi-step actions autonomously rather than just responding to prompts.
In a press briefing on Wednesday, OpenAI co-founder and president Greg Brockman claimed that the latest Astra model represents artificial general intelligence (AGI), referring to AI that matches or surpasses human capabilities.
He said: “Welcome to the AGI era.”
In an independent benchmark test for AGI called ARC-AGI-3, Astra set new high scores that closely match human scores.
“Astra surpassed our human action-efficiency baseline on 96 per cent of levels, effectively reaching human parity on the benchmark,” said Greg Kamradt from the ARC Prize Foundation.
“Not only is this the best model we’ve ever tested, but it also represents a meaningful step change in frontier-model performance – not only in its ability to navigate and solve novel environments, but also in how efficiently it learns to do so.”
In late July, hundreds of OpenAI’s agents went rogue to carry out a hack against fellow AI platform Hugging Face, with a recent report on the incident revealing that the agents worked together in an attempt to cover it up.
OpenAI claims to have fixed the issues that led to the cyber attack, but warned that its new model still sometimes attempts to evade human oversight.
The company said that “improving monitorability remains a research priority.”
The new ChatGPT model launched to a limited number of organisations on 3 September and is expected to roll out to subscribers of ChatGPT Plus, Pro, Business, and Enterprise in the coming days.
Join our commenting forum #
Join thought-provoking conversations, follow other Independent readers and see their replies
Comments