AlphaGo Creator Raises $1B for AI Without LLMs
The creator of AlphaGo has raised a record $1 billion for a new venture aimed at building artificial intelligence without relying on large language models (LLMs). This significant funding round signals expanding investor appetite for alternative AI architectures, such as agentic platforms and specialized vertical models.
- The new venture is Adept AI, co-founded by David Luan, Ashish Vaswani, and Niki Parmar, all of whom were instrumental in developing the Transformer architecture at Google. The company has raised a total of $415 million, achieving a $1 billion valuation after its $350 million Series B round. - Key investors in Adept AI's Series B funding include General Catalyst and Spark Capital, with participation from Greylock, Addition, and Nvidia. This syndicate of high-profile investors highlights significant venture capital interest in AI approaches that are not solely dependent on large language models. - Adept's core technology, ACT-1 (Action Transformer-1), is designed to translate natural language commands into actions within existing software applications, essentially acting as an AI teammate that can execute complex workflows. This "text-to-action" model works through a Chrome extension, allowing it to interact with various web-based tools and APIs. - AlphaGo's creator, Mustafa Suleyman, is now the CEO of Microsoft AI and is focused on a "human-first" approach to superintelligence. While he previously co-founded Inflection AI, which raised $1.3 billion, his new role at Microsoft involves leading the development of consumer AI products like Copilot. - The move toward agentic and vertical AI reflects a market trend where specialized agents are outperforming general-purpose models in specific industries like real estate and healthcare. These vertical AI agents are trained on domain-specific data to automate complex workflows, such as property valuation or lease analysis, with high accuracy. - Alternatives to LLMs for specific tasks include encoder-only models like BERT for text classification and semantic search, which can be faster and more cost-effective. Other approaches gaining traction are hybrid AI, which combines deep learning with symbolic reasoning for better transparency, and state-space models (SSMs) for more efficient processing of long data sequences. - In the real estate sector, vertical AI is already being applied to automate property management tasks, predict property values with over 90% accuracy, and enhance lead generation through smart chatbots. Companies like Zillow and Realtor.com use these specialized AI models to improve valuation accuracy and personalize user recommendations. - The broader agentic AI platform market includes companies like Kore.ai and Moveworks, which provide tools for enterprises to build and orchestrate multiple specialized AI agents for tasks in IT, HR, and customer service. These platforms enable AI to move beyond analysis to autonomous action and decision-making across various business systems.
