伦敦AI公司Inherent由谷歌DeepMind校友创办，该公司宣布其AI Agent产品Faraday在独立复制已发表科学论文成果的任务中表现超群1。值得注意的是，Faraday仅使用27亿参数的Qwen 3.6模型，却超越了Anthropic的Claude Opus 4.8和OpenAI的GPT-5.5等参数规模更大的前沿模型1。
Inherent已完成5000万美元种子轮融资1。该公司通过强化学习方法培养AI的"研究品味"，计划到年底将员工数从12人扩充至约20至25人1。
Inherent, a London-based AI company founded by former Google DeepMind researchers, announced that its AI agent Faraday has surpassed larger frontier models from Anthropic and OpenAI at independently replicating published scientific research findings.1 The system accomplished this using only Qwen 3.6, a model with 2.7 billion parameters, while competing against Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5.1
The company has secured $50 million in seed funding and plans to expand its workforce from 12 employees to between 20 and 25 people by year's end.1 Inherent's approach employs reinforcement learning techniques to cultivate what the company describes as an AI "research taste," enabling the agent to develop stronger performance on complex scientific tasks.1 The startup, headquartered in London's King's Cross district, was established by four co-founders with connections to DeepMind.1
评论
还没有评论，欢迎留下第一条。