According to Forbes, former xAI researcher Eric Zelikman is raising $1 billion for a new startup called Humans& that will train AI models to be better at collaborating with humans, with the company in talks for a $5 billion valuation. The founding team includes early Google employee Georges Harik, who was Google’s 7th employee and co-creator of Adwords and Adsense, along with researchers from Meta, Anthropic, OpenAI, and DeepMind. Zelikman, who wrote what he calls “the first paper to train language models to reason in natural language,” has emphasized that Humans& will focus on AI that understands human preferences and interests rather than replacing human capabilities. The company has told investors its new training paradigm will require more computing power than current AI training strategies, positioning it among a small group of well-funded labs like Thinking Machine Labs and Safe Superintelligence that are raising significant capital before product release. This ambitious funding round signals a potential shift in AI development priorities.
Table of Contents
The Philosophical Divide in AI Development
The emergence of Humans& represents a fundamental philosophical split in the artificial intelligence industry that has been brewing for years. While most major labs focus on creating increasingly autonomous systems that can operate independently, Humans& is betting that the real breakthrough will come from systems designed specifically to enhance human capabilities rather than replace them. This approach acknowledges that human intelligence operates differently from artificial intelligence – we excel at contextual understanding, emotional intelligence, and collaborative problem-solving in ways that current AI systems struggle to replicate. By focusing on collaboration rather than replacement, Humans& is essentially arguing that the most valuable AI systems won’t be those that work alone, but those that work seamlessly with human teams.
The Technical Hurdles of Collaborative AI
Building AI systems that genuinely collaborate with humans presents unique technical challenges that go beyond current model training approaches. Traditional AI training focuses on optimizing for specific tasks or benchmarks, but collaborative AI requires understanding human intent, preferences, and working styles in real-time. This likely involves developing new architectures for memory and context retention that can track individual user patterns across extended interactions. The company’s claim that their approach requires “more compute than current AI training strategies” suggests they’re planning something fundamentally different from the transformer-based architectures dominating today’s landscape. Given Zelikman’s background in reasoning models and his team’s experience at companies like Anthropic and OpenAI, they may be developing hybrid systems that combine different AI paradigms to achieve this collaborative capability.
The High-Stakes Bet on Unproven Technology
A $1 billion funding round for a pre-product startup represents an enormous gamble, even in today’s AI-crazed investment landscape. The investor who passed on the round citing the “too big of a number” highlights the legitimate concern about valuation inflation in the AI space. While the founding team’s credentials are impressive – spanning Google’s early days, xAI, and leading AI research labs – they’re essentially asking investors to bet on an unproven technical approach in a field where even well-established methods face significant limitations. The company will need to demonstrate rapid progress to justify this valuation, particularly as they’re competing for talent and resources with well-funded incumbents like Meta Platforms and other major tech companies that have vastly larger budgets and infrastructure.
What This Means for the AI Ecosystem
If Humans& succeeds in raising this funding round, it could trigger a wave of investment in “collaborative AI” startups, much like we’ve seen with autonomous AI companies. This would represent a significant diversification of the AI investment landscape beyond the current focus on either foundation models or application-layer companies. The emphasis on human-AI collaboration also addresses growing public and regulatory concerns about AI replacing human jobs, potentially making this approach more palatable to enterprises worried about workforce disruption. However, the company faces the challenge of defining and measuring “collaboration” in ways that convince both investors and future customers that their approach delivers tangible business value beyond what existing AI systems can provide.
The Road Ahead for Humans&
The success of Humans& will depend on their ability to translate Zelikman’s vision of “all IQ and no EQ” into practical technology that delivers measurable improvements in human-AI teamwork. Given the team’s background in reasoning models and their stated focus on systems that “remember and react to a person’s preferences and interests,” we can expect their initial products to target knowledge work and creative collaboration scenarios. The involvement of founders with backgrounds at companies spanning Google, Meta, and Anthropic suggests they understand both the technical and product challenges ahead. However, with no announced timeline for product release and the enormous computational requirements they’ve acknowledged, investors should expect a long development cycle before seeing whether this $1 billion bet on human-centered AI pays off.
