Leading AI researcher Eric Zelikman is raising $1 billion to build AI models with emotional intelligence (www.businessinsider.com)

Top AI researcher Eric Zelikman is raising roughly $1 billion at a reported $4 billion valuation to launch Humans&, a new startup focused on building language models with emotional intelligence and long-term user understanding, according to sources. The milestone round — still in progress — comes amid a wave of outsized seed rounds for labs helmed by high-profile researchers and entrepreneurs. Zelikman, a Stanford PhD student who left xAI in September and co-authored a high-profile paper on getting language models to “think before speaking,” says current models treat each conversational turn as an isolated game and lack awareness of longer-term user goals and values. Humans& intends to change the objective functions and training targets of models to emphasize empathy, user modeling and collaborative reasoning: learning from individual users, inferring their ambitions and values, and coordinating across groups. Technically this implies shifts toward personalized, stateful models that maintain longer context, optimize for human-aligned objectives, and potentially integrate multi-user signals for better collaboration. If successful, that could improve alignment, personalization and use in complex, long-horizon tasks (from advice to scientific collaboration), but it also raises questions about privacy, evaluation metrics for “empathy,” and whether huge early valuations match product readiness.