Additional reporting by Iain Martin and Richard Nieva.
Humans will train AI models to work higher with humans.
dpa/picture Alliance via Getty Images
Former xAI researcher Eric Zelikman is raising $1 billion for a brand new startup called Humans& that teaches AI models to work higher with humans, six sources said Forbes. The company is in talks for a $5 billion valuation, sources added.
Other founders include early Google worker Georges Harik and researchers who worked at Meta, Anthropic, OpenAI and DeepMind, sources added.
Humans& declined to comment.
While it’s unclear what product the corporate plans to develop, Humans& has told investors that it’s working on a brand new strategy to train models to recollect and reply to an individual’s likes and interests in order that AI can empower humans, unlike other AI labs that give attention to how well AI can replace humans.
“I personally feel very strongly that if we develop models that are really good at working with large groups of people, that are really good at understanding different people’s goals, ambitions and values, understanding different people’s weaknesses, and how to coordinate with these large groups of people to make everyone more effective, we’re much more likely to solve them,” Zelikman told investor Sarah Guo on the No Priors podcast in early October.
The company has told investors that its recent training paradigm would require more computing power than current AI training strategies, a source said.
Given the large sum of money and computing required to construct these powerful systems, fewer startups try recent methods of coaching models, opting as a substitute to develop applications based on AI models like GPT-5, Claude Opus 4.1 and Grok 4. If the fundraising closes, Humans& would join a handful of labs which have raised significant sums before releasing a product, reminiscent of Mira Murati’s Thinking Machine Labs and Ilya Sutskever’s Safe Superintelligence.
An investor who passed on the round said the round size was “too large a number” for an early-stage company.
Do you’re employed at xAI and do you’ve got a tip for us? Contact reporters Anna Tong at atong@forbes.com, Rashi Shrivastava at rshrivastava@forbes.com or rashi.17 on Signal and Iain Martin at iain.martin@forbes.com or 646-739-6427.
Zelikman, a doctoral student in computer science at Stanford University, wrote, in his words, “the first paper that trained language models to reason in natural language.” website. OpenAI’s groundbreaking “o” series is an example of this kind of reasoning model. He then worked on pre-training data collection (data utilized in the initial phase of coaching models), reasoning and agent infrastructure at xAI.
Georges Harik was the seventh worker at Google, co-developer of Adwords and Adsense, and is, in accordance with him, a startup investor LinkedIn.
Zelikman’s doctoral advisor Noah Goodman, a professor of computer science and psychology at Stanford University, can also be certainly one of the founders. According to him, Goodman worked on the post-training team for Gemini LinkedIn.
A fourth founder, Andi Peng, worked at Anthropic in post-training and behavioral reinforcement learning LinkedIn.
Another member of the founding team, Ray Ramadorai, worked on system design for Microsoft’s large data centers LinkedIn.
“If we invest everything in autonomy and nothing in collaboration, only IQ and no EQ, the future will be colder,” Zelikman wrote on his blog website.
MORE FROM FORBES
