More Gemini Features Coming to Google TV
Google TV integrates Gemini AI for image/video generation, enhances Google Photos, and adds a dedicated YouTube Shorts feed to the homepage.
Read moreNumenta has been studying the brain for 17 years and now finally has a product that promises to improve AI efficiency by 100 times.
The company, based in Redwood City, California, was founded by computer pioneers Jeff Hawkins and Donna Dubinsky. They are launching Numenta Intelligent Computing Platform (NuPIC), a neuroscience-based AI commercial solution.
The product is based on theories about the brain and intelligence, which Hawkins wrote about in his 2021 book "A Thousand Brains."
Interestingly, Numenta has partnered with game startup Gallium Studios, founded by game pioneers Will Wright (co-founder of The Sims) and Lauren Elliott (co-creator of Where in the World is Carmen Sandiego). Gallium Studios chose Numenta as their AI partner because they face fundamental challenges in integrating AI into games and prioritizing user trust and privacy.
With NuPIC, Gallium Studios can run large language models (LLMs) on CPUs with high performance and use both generative and non-generative models as needed. With complete control over models and data, Gallium Studios expects to develop simulated AI players that can learn, adapt, and exhibit intelligent behavior.
A New Software Platform
NuPIC utilizes Numenta's unique architecture, data structures, and algorithms to efficiently deploy LLMs on CPUs. This groundbreaking platform represents a significant milestone in the field of AI by providing disruptive performance, cost reduction, and crucial privacy, security, and control features. Importantly, NuPIC aims to provide convenience for developers and software engineers without requiring in-depth knowledge of deep learning, said Numenta CEO Subutai Ahmad in an interview with VentureBeat.
Most LLMs rely on graphics processing units (GPUs), which is one of the reasons why GPU manufacturers like Nvidia have become AI giants over the years. However, Numenta chose to collaborate with x86 processor manufacturer Intel instead of GPUs because GPUs have a less flexible programming model compared to CPUs. The idea is to offload most of the processing work to CPUs to reduce the cost of LLMs.
"We realized that we are in a chaotic period of AI. Everyone wants to benefit from it, but not everyone knows where to start or how to achieve the performance they need to put LLMs into production," said Ahmad. "As the only platform based on the theory of a thousand brains intelligence, NuPIC provides the ideal performance result of elevating CPUs to run LLMs. With our optimized inference servers, model library, and training modules, you can choose the right model for your unique business needs, fine-tune your data, and run models on CPUs with high throughput and low latency, much faster than on Nvidia A100 GPUs—all with the highest level of security and privacy."
In addition, NuPIC offers security and privacy protection for enterprises. Its features include consistent high throughput and low latency inference using only CPUs, eliminating the need for complex and expensive GPU infrastructure.
Unlike alternative solutions that require sending internal data to external software-as-a-service (SaaS) services, NuPIC runs entirely within the client's infrastructure and can be deployed on-premises or on dedicated clouds from major cloud providers. This approach ensures full control over data and models, ensures consistent behavior, reduces costs, and enhances data compliance.
NuPIC's flexible model library also provides a range of ready-to-use models, including BERT and GPTs. Customers can optimize for accuracy or speed and create customized versions of existing models that suit their needs.
Moreover, NuPIC enables customers to rapidly prototype LLM-based solutions without extensive knowledge of machine learning. With support from a dedicated team of AI experts, NuPIC simplifies seamless deployment of LLMs in production. Delivered as Docker containers, customers can iterate and scale them using standard MLOps tools and processes.
These unique features bring significant business advantages, according to Ahmad. NuPIC allows customers to leverage the power of LLMs on accessible CPUs, achieving remarkable throughput and latency improvements, running faster on Intel's 4th Gen Xeon Scalable processors and delivering 10 to 100 times performance improvement.
NuPIC enables customers to choose the right LLM, fine-tune it with custom data, easily scale, and handle larger models without expanding the budget. Most importantly, NuPIC empowers organizations to have complete control over their data, ensuring privacy and building trust.
Numenta currently provides access to NuPIC only to a limited number of enterprise customers. The company has around 20 employees and is privately funded through internal and external sources.
A Long Journey
Since 2005, Ahmad has been conducting technical research with Hawkins. The initial idea was to understand the brain and its efficient operation principles and then mimic those abilities in computer science. Many have attempted this idea, but IBM, for example, failed with its brain-inspired research. However, Hawkins proposed a unique theory.
Initially, Hawkins founded the Redwood Neuroscience Institute and published a book called "On Intelligence" in 2004.
In the first part, Hawkins pointed out that certain parts of the brain, such as memory, are related to a hierarchical structure, particularly a temporal hierarchy. He noted that you can remember things that happen in chronological order, which explains why you can easily remember music. The brain is like a prediction machine, learning from the past and making guesses about the future.
Turning Theory into Practice
Now, Hawkins believes that there may be 100,000 brain regions, or cortical columns, operating as independent brains within the overall system. Different cortical columns collaborate when thinking.
"We always felt that there were more fundamental things to learn from neuroscience," said Ahmad. "We now have a complete understanding of how to achieve intelligence in the neocortex. The field of neuroscience has exploded in the past 20 to 30 years. We think it's time to take that understanding and turn it into real algorithms and implement them in AI systems."
It is a software application that can run on any Intel CPU-compatible system. This means it can run on Intel and AMD CPUs, but currently not on Arm-based CPUs. Intel has validated the feasibility of this technology, Ahmad said.
"We've been in this field for a long time, deeply studying neuroscience, really trying to understand how the brain works," Ahmad said. "So, Jeff published a book called 'A Thousand Brains,' which came out two years ago. It really encapsulates our research and what we've learned from neuroscience."
Bill Gates named the book one of the top five books of 2021. Numenta investigated how this theory could impact practical AI systems.
"It turns out that we can take this knowledge from neuroscience and make large language models (LLMs) or GPT models 100 times more efficient," said Ahmad.
For example, if you show a picture of a cat to a person, they immediately learn that it's a cat. For an AI model, you have to show it thousands of pictures of cats before it can recognize a cat. Numenta started building algorithms for these shortcuts.
"The key is learning how to map the knowledge of the brain into existing hardware systems from an engineering perspective," said Ahmad. "Once we found a way to map it onto hardware systems, we can run it without having to build our own brain hardware."
The company has proven its commercial viability and is generating revenue.
He pointed out that the brain is very efficient, requiring only about 20 watts of power, while deep learning systems require a lot of GPUs. By switching to CPUs, processing can be more efficient, Ahmad said.
"We think this is a watershed moment," he said. "People can use it on regular servers and CPUs. They don't need to get special GPU systems. Once it runs on CPUs, you have a lot of flexibility, whereas with GPUs, it's hard to program them to be flexible."
He said you can run multiple models simultaneously without the need for batch operations. It also doesn't require running in cloud services, thus providing better privacy, security, and control.
Docker containers provide a simple way to run software without complex installation processes.
"Enterprises can save a lot of money with unmatched cost-effectiveness," he said.
Ahmad noted that the focus on CPUs makes sense because high-end GPUs are almost sold out due to the AI revolution, and it will be difficult to buy them for the next year or year and a half. And compared to GPUs, they are relatively less flexible and typically perform parallel computations of similar types.
"This is very important for us to be able to achieve these more innovative algorithms," Ahmad said. "The brain is not just doing a lot of matrix multiplications. It selectively decides what you want to compute, when to compute, and how to allocate computation because it's all metabolic energy in the brain. So, it has developed many clever strategies. But to do that, you need to be able to write these algorithms in a flexible way. So, CPUs are better than GPUs in this regard. CPUs are much more flexible than GPUs."
One trick the brain uses is avoiding computation rather than doing a lot of useless or redundant computations, Ahmad said.
"That's the idea we brought into transformers," he said.
To me, these sound a bit like using AI in computer graphics. Nvidia calls it DLSS (Deep Learning Super Sampling), where it assumes that a small patch of green in an image likely means the next pixel it renders will also be green and assumes the pixel will be green, thus skipping a lot of computations. This is an example of AI and graphics working together.
You can deploy existing LLMs on Numenta's optimized inference servers to run on CPUs. Then you can build applications on top of them. Additionally, Numenta has a training model, so you can fine-tune the model to make it more suitable for your application. As it is delivered as Docker containers, it can run on the client's infrastructure, such as Gallium Studios' own system.
Gallium Studios and the Future of Gaming with AI
This is exactly the future of gaming and AI, where you can find characters like yourself in Gallium Studios' game, called "Proxi."
"Our latest game, 'Proxi,' is a vast interactive world composed of your personal memories and associations. We chose Numenta because of the fundamental challenges we face in integrating AI into games. We want to provide the best experience for players while ensuring we don't compromise their trust and privacy. We can run LLMs on CPUs with incredible performance and use both generative and non-generative models as needed. With everything deployed locally, we have full control over models and data. Over time, Numenta's cutting-edge neuroscience-driven research will enable us to build simulated AI players that exhibit continuous learning, adaptation, and truly intelligent behavior," said Gallium Studios CEO Lauren Elliott in a statement.
These proxis simulate you to some extent. To run efficiently on CPUs, you need to be able to run a large number of models simultaneously, which is challenging on GPUs.
"Our product fits their game very well. I personally think it fits many different games that want to incorporate AI," Ahmad said. "Depending on the exact model, they might see performance improvements of two orders of magnitude and a huge difference in cost-effectiveness."
Google TV integrates Gemini AI for image/video generation, enhances Google Photos, and adds a dedicated YouTube Shorts feed to the homepage.
Read moreAlphabet Q1 paid subscribers reached 350M, fueled by YouTube & Google One. YouTube ad revenue slightly missed at $9.88B amid Premium growth. Total Q1 revenue hit $109.9B, beating forecasts.
Read moreAnthropic releases Claude connectors for Adobe, Blender, and Ableton, enabling AI workflow automation and app control. It also sponsors the Blender Development Foundation.
Read moreApple introduces an App Store model splitting 12-month plans into monthly installments. On iOS 26.4+, it lowers upfront costs without reducing total price, adding billing tracking.
Read moreGoogle launches Aadhaar digital IDs in Wallet for India, enabling secure QR/NFC verification. A phased partner rollout targets key sectors, while passport-based credentials expand globally.
Read moreYouTube tests "Ask YouTube," an AI tool blending text and video for complex queries. Initially for US Premium users, it supports interactive follow-ups, expanding Google's AI search.
Read moreGoogle grants US DoD classified AI access, joining OpenAI and xAI in defense deals amid Anthropic's dispute. Contractual safeguards against surveillance and lethal weapons remain unenforceable.
Read moreMistral AI launches Workflows, a Temporal-powered orchestration platform for secure, Python-driven enterprise AI automation featuring human-in-the-loop approvals.
Read moreCustom subliminal audio maker

AI Social Media Manager

AI Search Visibility

Workout and Calorie Tracker
Stock research platform

AI interview assistant

Scientific figure generator

AI courses with real code