Latest news with #Collective-1
Forbes
3 days ago
- Business
- Forbes
Shepherding Data To Its Eventual Destination
What does it mean to engineer data transfers in the age of AI? More than a few top professionals would say that this sort of thing is even more important now than it had been in the past. After all, artificial intelligence works on the principle of taking in information and processing it – in ways that are eerily similar to what happens in our human brains. So the information itself is valuable, and the processes are valuable, too. Also, modern advances in large language models have given us different ways to look at data transfer. I'll get into some of that with thoughts on three lightning talks given at Imagination in Action in April, in the Lightning Talks section of the wide-ranging program. Unlocking Siloed Data The first such presentation was from William Lindskog Munzing who was talking about an application called Flower, suggesting that traditionally, data has been trapped in siloes. The goal, he said, is to move AI to where the data is. That's much easier in times when work on quantizing and foundation models and lower-bit systems has ushered in edge AI – the ability to locate the AI where the data already is, rather than porting it to centralized data centers. With that in mind, the Flower community numbering some 5800 developers, with 2000 active projects, is working on what Munzing calls the 'Collective-1' user-owned platform. 'What we believe in is that data should stay where it originates,' he said. 'It's … never transmitted. It stays in your device, in your car, whatever it is, or in the hospitals.' The ISO-certified project, he added, is also versatile. 'We have done a lot of things in the deployment runtime,' Munzing explained. 'So it's very easy for you now to run your experiment on CPUs, GPUs, and then scale it to actual real world deployment if you want to add secure mechanisms, authentication and much more.' AI for Leads The next talk came from Marco Cello, who worked on a project called Meshify. He explained that data suggests small to medium sized enterprises (SMEs) are on average about 50% as efficient as corporations, and that collectively, SMEs lose up to $500 billion in revenue from poor lead management alone. As a solution, Meshify, he explained, will scour a professional's inbox, follow up, and provide them with automated CRM insights. I thought it was also interesting that this project uses the NANDA decentralized network idea pioneered at MIT by some colleagues of mine. Actually, so does Flower, which indicates the initiative to create a functional web protocol for AI is picking up steam. Regulation and Control The third presentation came from Peeyush Aggarwal at Deloitte, where he talked about dimensions of change in the AI era. This was different, because Aggarwal wasn't promoting a particular startup or product. Instead, he was talking about cycles – a cycle from assistive to augmentation to automation, and priorities for AI development, including: 'Human strengths need to be amplified,' Aggarwal said. 'How do you take a banker and train them to do cognitive, as well as decision making, when people are designed to think in a straight line, right? And it's about that ability to negotiate, to be able to take decisions ethically in an environment, (this) is the most important aspect.' Referencing a need to control the culture of change in the banking industry, Aggarwal went over many aspects of analyzing AI activities from a boardroom perspective, citing a gradual change and evolution from classic to digital banking, and then to intelligent banking. At the end of his presentation, he went into a sort of an interesting philosophical path, talking about the management of human and AI agents. 'The most important part is, when you bring human and AI agents together, is, do you control the AI, or do you control the decision of AI and humans coming together?' he asked. 'That's the most important aspect. What's the use case? When a regulator looks at a use case, he asks the question, 'Can I repeat that use case? Can I repeat the question that is being asked?' And if you can't, you can't really approve their use case.' In other words, managers who are managing human and AI agents have a different role than those who are managing only humans themselves. You're managing the intersection of humans with technology. How does this work? I thought these were some interesting eye-openers in a time when we're trying to adjust to a rapidly changing target in terms of technology use. Stay tuned.
WIRED
30-04-2025
- Business
- WIRED
These Startups Are Building Advanced AI Models Without Data Centers
Apr 30, 2025 12:00 PM A new crowd-trained way to develop LLMs over the internet could shake up the AI industry with a giant 100 billion-parameter model later this year. Photo-Illustration:Researchers have trained a new kind of large language model (LLM) using GPUs dotted across the world and fed private as well as public data—a move that suggests that the dominant way of building artificial intelligence could be disrupted. Flower AI and Vana, two startups pursuing unconventional approaches to building AI, worked together to create the new model, called Collective-1. Flower created techniques that allow training to be spread across hundreds of computers connected over the internet. The company's technology is already used by some firms to train AI models without needing to pool compute resources or data. Vana provided sources of data including private messages from X, Reddit, and Telegram. Collective-1 is small by modern standards, with 7 billion parameters—values that combine to give the model its abilities—compared to hundreds of billions for today's most advanced models, such as those that power programs like ChatGPT, Claude, and Gemini. Nic Lane, a computer scientist at the University of Cambridge and cofounder of Flower AI, says that the distributed approach promises to scale far beyond the size of Collective-1. Lane adds that Flower AI is partway through training a model with 30 billion parameters using conventional data, and plans to train another model with 100 billion parameters—close to the size offered by industry leaders—later this year. 'It could really change the way everyone thinks about AI, so we're chasing this pretty hard,' Lane says. He says the startup is also incorporating images and audio into training to create multimodal models. Distributed model-building could also unsettle the power dynamics that have shaped the AI industry. AI companies currently build their models by combining vast amounts of training data with huge quantities of compute concentrated inside datacenters stuffed with advanced GPUs that are networked together using super-fast fiber-optic cables. They also rely heavily on datasets created by scraping publicly accessible—although sometimes copyrighted—material, including websites and books. The approach means that only the richest companies, and nations with access to large quantities of the most powerful chips, can feasibly develop the most powerful and valuable models. Even open source models, like Meta's Llama and R1 from DeepSeek, are built by companies with access to large datacenters. Distributed approaches could make it possible for smaller companies and universities to build advanced AI by pooling disparate resources together. Or it could allow countries that lack conventional infrastructure to network together several datacenters to build a more powerful model. Lane believes that the AI industry will increasingly look towards new methods that allow training to break out of individual datacenters. The distributed approach 'allows you to scale compute much more elegantly than the datacenter model,' he says. Helen Toner, an expert on AI governance at the Center for Security and Emerging Technology, says Flower AI's approach is 'interesting and potentially very relevant' to AI competition and governance. 'It will probably continue to struggle to keep up with the frontier, but could be an interesting fast-follower approach,' Toner says. Divide and Conquer Distributed AI training involves rethinking the way calculations used to build powerful AI systems are divided up. Creating an LLM involves feeding huge amounts of text into a model that adjusts its parameters in order to produce useful responses to a prompt. Inside a datacenter the training process is divided up so that parts can be run on different GPUs, and then periodically consolidated into a single, master model. The new approach allows the work normally done inside a large datacenter to be performed on hardware that may be many miles away and connected over a relatively slow or variable internet connection. Some big players are also exploring distributed learning. Last year, researchers at Google demonstrated a new scheme for dividing and consolidating computations called DIstributed PAth COmposition (DiPaCo) that enables more efficient distributed learning. To build Collective-1 and other LLMs, Lane and academic collaborators in the UK and China developed a new tool called Photon that makes distributed training more efficient. Photon improves upon Google's approach, Lane says, with a more efficient approach to representing the data in a model and a more efficient scheme for sharing and consolidating training. The process is slower than conventional training but is more flexible, allowing new hardware to be added to ramp up training, Lane says. Photon was developed in collaboration with researchers at Beijing University of Posts and Telecommunications and Zhejiang University in China. The group released the tool under an open source license last month, allowing anyone to make use of the approach. Flower AI's partner in the effort to build Collective-1, Vana, is developing new ways for users to share personal data with AI builders. Vana's software allows users to contribute private data from platforms like X and Reddit to training a large language model, and potentially specify what kind of end uses are permitted or even benefit financially from their contributions. Anna Kazlauskas, cofounder of Vana, says the idea is to make untapped data available for AI training and also to give users more control over how their information is used for AI. 'This is data that isn't usually able to be included in AI models because it's not publicly available,' Kazlauskas says, 'and is the first time that data directly contributed by users is being used to train a foundation model, with users given ownership of the AI model their data creates.' Mirco Musolesi, a computer scientist at University College London, says a key benefit of the distributed approach to AI training is likely to be that it unlocks new kinds of data. 'Scaling this to frontier models would allow the AI industry to leverage vast amounts of decentralized and privacy-sensitive data, for example in health care and finance, for training without the risks associated with data centralization,' he says. What do you think of distributed machine learning? Would you contribute your data to a model like Collective-1? Send an email to hello@ or comment below to let me know.
