For Israel to win in AI it must play a different game
Israel's success in AI depends on shifting focus beyond just scaling models, according to insights from Safe Superintelligence Inc., a company co-founded by Ilya Sutskever. The company is exploring alternative approaches, including better reasoning loops, tools, and feedback mechanisms.
Israel's AI success is being reevaluated as the industry faces limitations in scaling models. The current approach, focused on increasing model size and training data, may not be sufficient for achieving superintelligence. Researchers have found that public human-generated text is finite and may be exhausted between 2026 and 2032. Moreover, next-token prediction is not equivalent to judgment, and models struggle with tasks like arithmetic and factual lookup. Safe Superintelligence Inc., co-founded by Ilya Sutskever, Daniel Gross, and Daniel Levy, is exploring alternative approaches. The company, with offices in Palo Alto and Tel Aviv, is assembling a lean team focused on superintelligence. Recent results from systems that add structure, search, and feedback, such as AlphaGeometry and FunSearch, suggest that these approaches may be crucial for achieving superintelligence.
This content was automatically generated and/or translated by AI. It may contain inaccuracies. Please refer to the original sources for verification.