Institutional development · Digital transformation, data and AI · Israel
Tel Aviv–Stanford preprint reports AI gaining predictive skills from self-generated data
Researchers from Tel Aviv University and Stanford University report in an arXiv preprint that AI models can develop general predictive skills using only training data generated within the system.
Why it matters: AI researchers can examine, in the arXiv preprint, a training method that Tel Aviv University says relies only on data the system generates itself.
Tel Aviv University said a study by its researchers and Stanford University colleagues shows AI can develop general predictive skills from self-generated training data alone.
According to the university, the method has two models start from scratch and learn together, with the first producing short computer programs that output sequences of symbols.
The university said the work was supervised by Dr Yoav Levine of its Blavatnik School of Computer Science as part of Dr Kfir Dolev's postdoctoral research.
The study has been posted as a preprint on arXiv.
Editor’s view
Sarah Williamson, Teaching, Digital and AI Editor
The university presents the study as challenging the assumption that AI must learn from large volumes of real-world data; for now the findings are available to AI researchers as an arXiv preprint.
