Meta AI Introduces DeepConf: First AI Method to Achieve 99.9% on AIME 2025 with Open-Source Models Using GPT-OSS-120B
Giant language fashions (LLMs) have reshaped AI reasoning, with parallel pondering and self-consistency strategies usually cited as pivotal advances. Nonetheless, these strategies face a elementary trade-off: sampling a number of reasoning paths boosts accuracy however at a steep computational price. A group of researchers from Meta AI and UCSD introduce Deep Suppose with Confidence (DeepConf),…
