Sakana AI: We’re a Tokyo-based R&D company on a mission to create a cool foundation model inspired by nature’s smarts. Our goal is to mix top-notch tech with the clever ways of the natural world. By using the principles of natural intelligence.
We aim to make a super innovative model that not only boosts technology but also clicks with the natural smarts we see around us. Come join us on this exciting journey as we shake things up and add to the evolution of artificial intelligence and more.
Sakana AI
Artificial intelligence has come a long way, from AlphaGo’s victories to language models like GPT-3 mastering natural language. However, the current method of training large models with massive datasets in supervised learning is hitting its limits. We need more efficient approaches to propel AI forward.
Meet Sakana AI, a Tokyo-based startup led by former Google AI researchers David Ha and Llion Jones. Inspired by nature, specifically schools of fish and flocks of birds, Sakana AI aims to create AI models that work well with less data.
If successful, this approach could result in smarter and more agile AI agents, behaving much like natural organisms. It’s a promising step towards a new era in artificial intelligence.
The Promise of Foundation Models
In AI, a novel paradigm known as basis models has surfaced in recent years. Large datasets are used to train these models so they can learn general skills that can be applied to a range of downstream applications.
Text generating examples include natural language models such as BERT and GPT-3. By eliminating the need to train new models for each new challenge, foundation models greatly increase the potential applications of AI.
Existing foundation models do, however, have certain drawbacks. Their training involves analyzing internet-scale datasets with thousands of GPUs, making it a computationally demanding process. The data they are trained on also limits their functionality.
We require more effective training procedures and the capacity to incorporate a variety of experiences in order to develop more capable, adaptable foundation models. Sakana AI is aiming to take advantage of this chance.
Smaller, Smarter Models
Sakana AI seeks to build AI foundations from the ground up as opposed to building massive foundation models. Their methodology draws inspiration from the emergent intelligence exhibited by natural flocks of birds and schools of fish.
Find a school of fish to watch. Individual fish exhibit basic behaviors in response to their surroundings, but as a school, they exhibit intricately coordinated movements that aid in avoiding predators and finding food. Similarly, aggregations of birds gather and swerve in accord, choreographed to a music none can hear.
Sakana The core tenet of artificial intelligence (AI) is that creatures with similar emergent group intelligence can be created. Sophisticated global behavior can arise from simple behavioral rules and localized sensory input. Much larger intelligence can emerge from the collective by concentrating on the fundamentals.
The Sakana AI researchers are investigating how AI agents may exchange experiences, pick up behaviors from one another, and work together to solve issues.
A small group of AI bots could demonstrate significantly greater intellect than the individuals by themselves with the correct algorithms. Compared to today’s massive models, this approach could greatly reduce data and computing requirements.
Bio-Inspired AI
Sakana AI is creating whole new kinds of artificial intelligence algorithms by delving even further into nature. Living things possess incredibly intricate talents that have developed over billions of years of adaption thanks to evolution.
Light is converted to chemical energy by plants. Processing sensory inputs is done by 100 trillion synapses in the human brain. Everything stems from basic incremental enhancements maintained by natural selection.
In order to recreate this phenomenon in AI, Sakana AI is leading the way with methods like computational evolution and bio-inspired design. Sakana AI will mimic evolution to increase intelligence by starting with the most fundamental building elements.
Instead of being directly programmed, the necessary characteristics gradually surface as a result of selective pressure. Beyond what human developers may imagine, computational evolution can produce innovative solutions.
Sakana AI aims to develop AI with greater adaptability, generalizability, and autonomy by modeling nature. When the scope of the problem or the data changes, AI models usually require human developers to monitor and update them.
On the other hand, natural creatures adapt dynamically without outside assistance. Artificial intelligence (AI) influenced by biology might advance constantly to meet new difficulties as the environment changes. This might quicken research in every AI-affected industry.
The Future of AI
It will take fundamental advancements in AI algorithms to realize these lofty goals of computational evolution and emergent collective intelligence. Fortunately, Sakana AI has already put together a stellar research team that includes both up-and-coming fresh talent and seasoned AI professionals.
Leading the company are Drs. David Ha and Llion Jones. Ha formerly led Google Research’s advances in automated inference and reasoning. At Google, Jones oversaw the development of conversation AI technologies.
Which are utilized on more than 2.5 billion devices globally. Elite scientists in the fields of evolutionary computation and machine learning have been hired by them from MIT, Stanford, Carnegie Mellon, and other prestigious academic institutes.
Prominent technological investors have also expressed interest in investing in Sakana AI. The startup stated that Lux Capital and Khosla Ventures, together with support from 500 Global and many other firms, are leading a $30 million early fundraising round.
The endorsement from prominent figures in the industry highlights the revolutionary possibilities of Sakana AI’s technology.
Sakana AI is now working on basic research, proof-of-concept trials, computational infrastructure, and collaborations with developers of consumer applications that want to test their innovations. Even if new technologies are still in the lab.
David Ha stated that he hopes to make an impact in five years. We hope that our developments will make it possible for a consumer AI assistant to grow with you over time. There will be exponential increases in functionality as hardware gets better.
We are still in the early days, but Sakana AI’s success over the next few years might establish the trajectory for the future of AI. Their biomimetic methodology is inspired by the creative genius of nature.
Our ability to create and use artificial intelligence would change if we could grasp collective emergence and computational evolution. Sakana AI is our civilization’s attempt to elevate all of humanity by reaching deeper reservoirs of intelligence.
Seed Funding Round
We are happy to announce that, with significant support from Khosla Ventures, we have successfully received $30 million in a seed funding round led by Lux Capital. These two venture capital firms have a long history of supporting long-term, R&D-focused businesses, particularly in AI.
Hugging Face and Runway are just two examples of the businesses Lux is renowned for its early support of, and Khosla was the first institutional investor in OpenAI.
As far as we are aware, Sakana AI is the first AI business established in Japan (or nearly the first in any industry) to receive seed funding from prestigious Silicon Valley venture capitalists.
Why did global venture capitalists finance Sakana AI?
“Everyone in AI is attempting to push the limits of scaling laws and chasing transformer architectures from yesterday.” Lux is constantly searching for the next big thing, from whom and where. The novel foundation models that the Sakana AI team is creating.
Which draw inspiration from complex adaptive systems and evolution, hold the key to our discovery. Sakana AI is special and significant since it is a Japanese company that initially concentrates on Asian markets. With one of the biggest economies.
A sizable market for enterprise software, and a tech-forward culture, Lux is quite optimistic about Japan. Asia will need and have a significant new AI trailblazer in a more complex world, and Japan, a strong democratic country and a valuable US ally, is well-positioned to be that person.
For reasons of national security as well as to more effectively communicate with local dialects, cultures, and values, the majority of sovereign nations will desire to have their own indigenous foundational models. We have backed this endeavor in India and are now doing it in Japan.
Furthermore, there is a constant need to deploy AI expertise for top-tier initiatives like Sakana AI, as it is dispersed globally. Sakana AI is well-positioned to take advantage of local talent and realize its full potential.
“We have long believed in the value of talent and markets around the world, having been among the first Silicon Valley venture firms to invest outside the US. Sakana AI is a prime example of not only the power of entrepreneurs with a global perspective.
But also the next generation of AI talent that will be drawn to the strongest firms from all over the world, including the United States. It is astounding how quickly David and Llion have been able to draw in the best AI talent on the planet.
As we begin this first stage of the AI technology change, we think Sakana AI has a great chance to have a significant impact on the industry, and we’re happy to support their expansion.
Founding Team
With this funding and support from top investors and companies, we’re now building a world-class AI lab in Japan! Our founding team members are from institutions such as Google and other unicorn AI Startups (Preferred Networks, Mercari):
- David Ha (Google Brain, Goldman Sachs)
- Llion Jones (Google Research, Transformer Co-Creator)
- Ren Ito (Mercari, Stability AI, Ministry of Foreign Affairs of Japan)
- Takuya Akiba (Preferred Networks, Stability AI)
- Yujin Tang (Google Brain, Google DeepMind)
- Robert Lange (Google DeepMind, TU Berlin)
- Tarin Clanuwat (Google Brain, Google DeepMind, Japan National Institute of Informatics)
- Makoto Shing (Rinna, Stability AI) (*)
- Qi Sun (LAION, Tokyo Institute of Technology)