Model releases and the enterprises that actually deploy them. Models coverage from Fathom Tech.
A San Francisco startup is rethinking how AI predicts human behavior, arguing that fine-tuned language models miss the mark and building a world model from scratch instead.
The startup, valued at $25 billion, has released Beam, an open-source large language model that matches the performance of much larger competitors while requiring significantly less computational power.
The Seattle AI research firm unveiled a new training framework that dramatically accelerates mixture-of-experts language models while reducing computational costs, enabling researchers without massive infrastructure to build models at scale.
Microsoft has released three new models in its MAI family, including its first real-time transcription engine and two text-to-speech variants, as the company works to reduce reliance on costly third-party AI providers.