Israeli Startup DoubleAI Challenges AI Giants With Novel Math Problem-Solving Approach
Translated & summarized from NEWSru Israel by baba
Israeli startup doubleAI, founded by Amnon Shashua, has advanced a 1988 mathematical problem by improving a graph coverage coefficient from 4/3 to 8/5 for only $2,000. The company uses an "Artificial Expert Intelligence" approach, focusing on deep dives into specific math areas rather than the broad, resource-intensive methods of AI giants like OpenAI. DoubleAI's system, Double Agent, generates hypotheses, verifies them, and learns from derivations, demonstrating a potentially more efficient path to AI-driven mathematical discovery. The startup is valued at over $1 billion.
The story in 6 lines · by baba
- Israeli startup doubleAI solved a 1988 math problem for $2,000 using focused AI.
- DoubleAI's approach contrasts with AI giants' costly, broad problem-solving strategies.
- The company focuses on "Artificial Expert Intelligence" (AEI) instead of general AI.
- Their system, Double Agent, improved a regular graph coverage coefficient to 8/5.
- DoubleAI's method is compared to AlphaZero's hypothesis generation and verification.
- The startup is valued at over $1 billion, challenging AI industry norms.
Israeli startup doubleAI, founded by Amnon Shashua, has presented an alternative strategy for artificial intelligence in mathematics, achieving a significant advance on a nearly 40-year-old problem for just $2,000. Unlike AI industry leaders such as OpenAI, which invest millions in supercomputers to tackle numerous problems simultaneously, doubleAI focuses on "Artificial Expert Intelligence" (AEI) rather than general intelligence (AGI).
The competitive landscape of AI-driven mathematical proofs, spurred by OpenAI, has revealed limitations in the mass-approach strategy. Large language models often produce overly complex and difficult-to-understand proofs. Furthermore, the mathematical community has noted a semantic gap: AI-generated proofs in Lean code for machines do not always align in meaning with human-readable text also generated by AI, complicating verification and leading to ambiguous results.
DoubleAI's system, Double Agent, demonstrated its effectiveness by improving the coverage coefficient for regular graphs from 4/3 to 8/5, a breakthrough in a problem that had seen no progress since 1988. The system generated a new algebraic construction, verified it using Lean, and produced a rigorous proof, all within a $2,000 budget.
The startup's efficiency stems from its focused approach, eschewing broad exploration for deep dives into specific mathematical domains. Double Agent operates similarly to DeepMind's AlphaZero, generating hypotheses, instant verification, and learning from successful formal derivations. Human researchers guide the search direction and set conditions, while the AI handles exhaustive computation.
With a valuation exceeding $1 billion, doubleAI's method suggests that focused, in-depth search combined with strict verification can outperform the brute-force methods employed by major AI companies.
Not the same event — other stories that share this one’s people, places, or theme: background, reactions, and follow-ups.
