Safe Superintelligence Inc., a prominent AI startup co-founded by former OpenAI chief scientist Ilya Sutskever and others, is reportedly seeking to raise new funding at a valuation of at least $20 billion.
According to Reuters, sources revealed that the company is in talks with both new and existing investors. Current investors in Safe Superintelligence Inc. (SSI) include Sequoia Capital, DST Global, Andreessen Horowitz, SV Angel, and NFDG. These investors helped secure $1 billion for the company last September, valuing it at approximately $5 billion at that time.
SSI now aims to raise funds at a valuation at least four times higher than its previous round, indicating strong investor confidence in its technology. The company was founded in June of last year with the goal of developing AI models capable of "superintelligence" while implementing safeguards against harmful outputs. The founding team includes Sutskever, AI researcher Daniel Levy, and Daniel Gross, a former Y Combinator partner who serves as the company's CEO and has led AI development efforts at Apple for several years.
Sutskever stated that SSI will focus solely on developing AI models with superintelligence and will not engage in any other activities, implying no immediate revenue generation. Few details have been disclosed about how SSI plans to build these models. According to Reuters, Sutskever mentioned the company would adopt "a new research direction" rather than continuing with current AI development methodologies.
Developers have traditionally enhanced the output quality of large language models (LLMs) by increasing the number of parameters along with hardware and training data volumes. At the recent NeurIPS machine learning conference, Sutskever suggested this approach may be nearing its limits. He noted that developers face challenges finding sufficient high-quality training data to further improve LLM performance.
"We've reached the peak of available data; there won't be more," Sutskever said. "We must work with what we already have. There's only one internet."
In 2012, Sutskever was part of the academic team that created AlexNet, one of the first modern computer vision models. This algorithm inspired extensive deep learning research and laid the groundwork for large language models. Following this project, Sutskever worked at Google LLC for several years before co-founding OpenAI in 2015.
Sutskever served as the chief scientist behind ChatGPT until last year and contributed to various OpenAI projects focused on optimizing reasoning in LLMs. Reasoning models were also a key topic in his recent NeurIPS keynote speech.
"The more a system reasons, the less predictable it becomes," Sutskever told attendees. "This is partly because even the best chess AIs are unpredictable to top human players."
SSI is not the only startup exploring new approaches to building AI models. In December, Liquid AI Inc. raised $250 million from investors to develop so-called liquid neural networks. The startup claims this algorithm can match the output quality of state-of-the-art LLMs using a fraction of the required hardware.
An AI model consists of artificial neurons, each a simple program performing part of the processing needed to generate responses to prompts. Each neuron includes components called weights and activation functions. Weights determine which data the AI considers when making decisions, while activation functions contain the code the neuron uses to analyze this data.
Standard LLM weights and activation functions remain unchanged after training. In contrast, liquid neural networks can reconfigure these components during inference, enabling the algorithm to continuously adjust its processing based on ingested data.
SSI is among several AI companies reportedly seeking new funding rounds. Last month, it was reported that OpenAI aims to complete a financing round valued at $40 billion, with a post-money valuation of $34 billion. Competitor Anthropic PBC is allegedly seeking to raise $2 billion at a $60 billion valuation.
