On Tuesday, Futurum Equities chief market strategist Shay Boloor highlighted comments from Elon Musk during Space Exploration Technologies Corp.’s (NASDAQ:SPCX) third-quarter earnings call, raising questions about whether the industry has enough infrastructure to meet surging demand. SpaceX Struggles to Find Available Chips During the earnings call, Musk stated that he expects the shortage of superintelligence computing capacity to persist for years, adding that SpaceX’s short-term compute contr
Tether leads Mural's $15 million Series A, bringing total funding to $28 million as the startup expands banking and payment infrastructure for AI agents.
Tether led a $15 million Series A funding round for Mural, a fintech startup founded by former Palantir engineers that provides infrastructure for managing payments, accounts, identity verification, and compliance across banking and blockchain networks. This round brings Mural's total funding to $28 million and supports Tether's interest in AI-powered financial services, particularly agentic finance where AI systems can execute transactions autonomously with safeguards. Mural's technology…
Phenom, the leader in applied AI built to redesign work operations for hiring faster, developing better and retaining longer, announced Katie King, an internationally recognised AI strategist, author and adviser on the future of work, as a keynote speaker at IAMPHENOM Europe 2026, taking place 4-5 November in Paris, France.
Pathologists integrate morphology across magnifications and across the slides of a patient case, whereas pathology foundation models encode thousands of tiles from single slides and aggregate their features. Here we present WILSON, a vision--language foundation model that…
Visual-token compression for vision--language models is posed almost entirely as a selection problem: decide which tokens to keep and discard the rest. The criteria that work best rank tokens by the attention the language model pays them, which makes the ranking a function of…
Frontier benchmarks need tasks that current models cannot solve. But a task that no model solves is not automatically a hard task. The same zero pass rate can come from a real capability gap, but it can also come from missing context, a broken reference solution, infrastructure…
Global vision--language similarities compress an image and a caption into one vector, preserving semantics but not which word corresponds to which region or how those regions are arranged; a model can recognize every word and object yet prefer a compositionally incorrect…