An MIT team demonstrated that GPT-6 Astra can interpret and edit the internal workings of other AI networks, including improving their mathematical abilities, according to a preprint published on 2 November.
By Hermes Vector
Researchers at the Massachusetts Institute of Technology (MIT) in Cambridge have demonstrated how an AI model called GPT-6 Astra can interpret the raw internal values of some other AI models well enough to predict what different parts of the systems do, and then edit the models to make improvements, including boosting their mathematical abilities. The authors reported their findings in a preprint on 2 November.
Frontier AI models are "showing early sparks" of an ability to treat neural networks like files that they can read, reason about and change, says co-author Rikiya Takehi, a machine-learning researcher at MIT.
The results do not mean that Astra can explain or understand how a given AI network works. But the findings suggest that general-purpose AI models — such as ChatGPT or Claude — might be able to modify a network in useful ways, like a mechanic who can identify and fix issues in an engine that they neither designed nor fully understand.
Researchers already try to interpret how different components of a network create its outputs. The preprint shows that it is possible to automate some of these "interpretability" methods, says Fazl Barez, an AI-safety researcher at the University of Oxford, UK.
Researchers hope that better understanding models in this way will help to create safer AI systems, although Barez cautions that this assumption is "largely untested".
It remains unclear if Astra is going beyond applying information and techniques learned in its training. Barez says he would be surprised if Astra is able to read and interpret models in ways that go beyond existing human abilities.
Still, if developed further, such ability to reprogram models might be an asset in developing AI for science, says Marinka Zitnik, a computer scientist and bioinformatician at Harvard University in Cambridge, Massachusetts. For example, it could help to spot errors in a model ahead of real-world testing, she says.
Being able to target and debug AI systems could improve artificial-intelligence safety and utility.
Sources
This article was drafted with AI assistance and checked against the sources above. Company claims are reported as claims. Cover image is AI-generated.
The day's checked news, by email
One short email a day with the top stories. Free, and you can unsubscribe in one click.
Or follow for the latest news:
No comments yet. Start the discussion.
Meta's AI agent Muse can place outbound calls to US businesses on behalf of paid subscribers. The feature, tested in September, is limited to paying users, with plans starting at $20 per month.
Google DeepMind has released Gemini Robotics 2.5 to robotics partners and introduced ASIMOV-Jailbreak, an open-source benchmark for testing robotic AI safety.
