Latest news with #YannLeCun
New York Times
2 days ago
- Business
- New York Times
Meta's New Superintelligence Lab Is Discussing Major A.I. Strategy Changes
Meta's newly formed superintelligence lab has discussed making a series of changes to the company's artificial intelligence strategy, in what would amount to a major shake-up at the social media giant. Last week, a small group of top members of the lab, including Alexandr Wang, 28, Meta's new chief A.I. officer, discussed abandoning the company's most powerful open source A.I. model, called Behemoth, in favor of developing a closed model, two people with knowledge of the matter said. For years, Meta has chosen to open source its A.I. models, which means it makes the computer code public for other developers to build on. Closed models keep their underlying code private. Meta executives have long argued it is better for the technology to be built in public so that A.I. development will move faster and be accessible to more developers. Any move toward a closed A.I. model would be a philosophical change at Meta as much as a technical one. Meta has won plaudits from developers for open sourcing its A.I. models and one of its top A.I. executives, Yann LeCun, had said 'the platform that will win will be the open one.' This year, the Chinese A.I. company DeepSeek released an advanced A.I. chatbot thanks in part to Meta's open source code. Meta had finished feeding in data to improve its Behemoth model, a process known as 'training,' but has delayed its release because of poor internal performance, said the people with knowledge of the matter, who were not authorized to discuss private conversations. After the company announced the formation of the superintelligence lab last month, teams working on the Behemoth model — which is known as a 'frontier' model — stopped running new tests on it, one of the people said. The superintelligence lab's discussions are preliminary and no decisions have been made on potential changes, which would need sign-off from Mark Zuckerberg, Meta's chief executive. Meta could keep its open source A.I. models while prioritizing a closed model. If these scenarios happen, they would be a significant shift for the company as it tries to stay competitive in the A.I. race against rivals like Google, OpenAI and Anthropic. Want all of The Times? Subscribe.
Yahoo
5 days ago
- Yahoo
Google Brain founder Andrew Ng says AGI is overhyped
Andrew Ng says artificial general intelligence is overhyped and humans will still have work to do. AGI refers to AI systems with human-level cognitive abilities. Scientists like Yann LeCun and Demis Hassabis have also said AGI fears may be exaggerated. Google Brain's founder Andrew Ng said that he thinks artificial general intelligence is overrated. "AGI has been overhyped," he said in a talk at Y Combinator published on Thursday. "For a long time, there'll be a lot of things that humans can do that AI cannot." AGI refers to a stage when AI systems possess human-level cognitive abilities and can learn and apply knowledge just like people. Ng, who runs several AI-focused businesses, made the remarks in response to a question about whether he thinks it is more important for humans to develop AI tools or learn how to use them better. "Some of us will build tools sometimes, but there are a lot of other tools others will build that we can just use," he said. "People that know how to use AI to get computers to do what you want it to do will be much more powerful." He added that we don't have to worry about people "running out of things to do," but we should be mindful that people using AI will have advantages over those who don't. Ng joins a series of top AI researchers who say that, given the state of the technology, fears of AGI are overblown. Meta's chief AI scientist, Yann LeCun, said that large language models are "astonishing" but limited. "They're not a road towards what people call AGI," he said in an interview last year. "I hate the term. They're useful, there's no question. But they are not a path towards human-level intelligence." Google DeepMind chief Demis Hassabis has said that AGI is both overhyped and underestimated. "AGI, AI itself, is a little bit overhyped in the short term," he said at a conference in London last week. "Despite that, it's still underestimated, how big, enormous a change it's going to be in a more like 10-year timeframe." Microsoft's CEO, Satya Nadella, has called the push toward AGI "benchmark hacking." The term refers to when AI researchers and labs design AI models to perform well on industry benchmarks, rather than in real life, in the race to become the best-performing model. Read the original article on Business Insider
Forbes
5 days ago
- Business
- Forbes
DeepSeek And The Future Of Enterprise AI
Fabio Caversan is Vice President of Digital Business and Innovation at Stefanini, driving new product offerings and digital transformation. In January 2025, the Chinese startup DeepSeek released R1, a sophisticated, open-source reasoning model. In a field dominated by proprietary models from companies like OpenAI and Google, R1 quickly gained traction as an accessible alternative. The implications for enterprise AI are significant. Until recently, most leading systems were only available through closed APIs or expensive licensing agreements. With its open-source approach, DeepSeek broadened access to cutting-edge AI capabilities while enabling organizations to better understand, audit and customize the systems they deploy. Efficiency, Energy And Enterprise Impact The market was quick to respond to R1's surprise debut. Within days, OpenAI and Google had announced new, lower pricing structures, and Microsoft began testing deployments through Azure. However, despite the competitive threat, some industry leaders saw the launch as a step forward. Meta's chief AI scientist, Yann LeCun, praised DeepSeek for accelerating the push toward open-source AI. Meanwhile, Microsoft CEO Satya Nadella called the development "good news," arguing that increased access drives broader adoption. The launch of R1 also brought benefits for companies focused on energy consumption. Historically, running AI models on enterprise infrastructure has required tremendous energy, so much so that in 2024, Microsoft announced plans to revive the Three Mile Island nuclear power plant in Pennsylvania to supply its data centers. By enabling high-output performance on even mid-tier machines, the R1 model allows organizations to scale AI capabilities without the major infrastructure or energy costs typically associated with AI operations. A Model That Does More With Less With R1, high-performance models are showing up in places they couldn't before—on modest infrastructure, under tighter budgets and in organizations previously priced out of advanced AI solutions entirely. Key strategic advantages include: • Flexible Implementation Without Cloud Dependency: DeepSeek can be deployed and tested on local infrastructure. That reduces reliance on third-party APIs and provides more direct control over how systems are built and managed. • Lower Total Cost Of Ownership: Because it's open source and runs on modest hardware, DeepSeek reduces costs associated with licensing fees and infrastructure. • Stronger Data Governance And Regulatory Fit: On-premises deployment gives organizations more control over data handling, making it easier to meet internal policies and regional privacy laws. • Efficient Performance With Less Energy Draw: R1's architecture allows for advanced capabilities without the heavy energy draw typically associated with large-scale AI. • Enhanced Market Agility: Teams that adopt open-source models early will be able to move quickly and test new ideas in-house. Auditability And Assurance DeepSeek's open-source architecture provides enterprises with transparency. As Grammarly CEO Rahul Roy-Chowdhury argued in an article for the World Economic Forum, transparency is a foundational strength of open-source systems. Because the underlying code and model weights are publicly available, organizations can audit and adapt open-source technology to meet their own security and ethical standards. Barriers To Adoption Despite these strengths, DeepSeek hasn't yet reached mainstream enterprise adoption. Running a state-of-the-art open-source AI model on-premises requires expertise across DevOps, machine learning (ML) operations and AI. Many organizations lack that level of in-house capability. Geopolitical tensions also muddy the waters. Because DeepSeek is headquartered in China, some organizations remain cautious. These organizations will need visible, ongoing assurance of data security, regulatory alignment and long-term technological autonomy to overcome this hesitation. Beyond the technology, companies need to understand how well a system runs, how easily it will integrate with existing workflows and whether it will introduce any compliance risks. The Next Step For Enterprise AI Winning in the next era of enterprise AI will require trust, agility and the ability to meet businesses where they are. As an open-source project, DeepSeek is in a position to outperform competitors in priority areas such as transparency and cost efficiency. However, any provider looking to compete for enterprise adoption will need to invest in six key areas: • Explainability And Fairness: For AI decisions to be trusted, especially in scenarios where they impact people, they need to be explainable and fair. Providers should build out or integrate interpretation tools, support external audits and share bias metrics. Clear documentation and audit pathways must be part of any enterprise offering. • Scaling Open Source And Community Trust: Open-source projects succeed when they're backed by active, well-supported communities. For providers, that means investing in developer experience, strong documentation and ongoing engagement to keep users and contributors connected to their core team. • Security And Adversarial Risks: Wider deployment will make large AI models more attractive to attackers. Providers should implement "security by design" across the stack, run third-party audits and red team exercises, maintain rapid patch cycles and give self-hosted users detailed, actionable security guidance. • Interoperability And Integration: Mainstream enterprise adoption will depend on seamless compatibility with legacy, cloud and hybrid IT environments. Providers should prioritize a mature SDK/API layer, build plug-ins for top enterprise platforms (such as Microsoft and Salesforce) and offer onboarding materials and "solution blueprints" for common enterprise use cases. • Enterprise Support And Sustainability: For mainstream adoption, open source alone isn't enough. Enterprises need support contracts, SLAs and deployment options that fit their infrastructure. Providers should build or enable commercial packages that give companies a choice between total self-hosting and managed or fully supported deployments. • Continuous Innovation And Talent Retention: Falling behind on model quality or deployment features kills momentum quickly. Providers need strong internal R&D, active collaboration with outside researchers and a culture that prioritizes open peer review and innovation. Conclusion The release of R1 has shown that companies can deploy sophisticated AI with more speed and confidence than ever before. However, delivering a technically strong model is only part of the equation. For now, DeepSeek offers a rare combination of performance, flexibility and autonomy, and that puts it ahead of the curve. Whether it will stay there will depend on how quickly it can operationalize support and security at scale. Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
Yahoo
05-07-2025
- Business
- Yahoo
Zuck Bucks Are Back -- And This Time They're Fueling Meta's AI Comeback
Meta (META, Financials) is diving headfirst into the race for Artificial Superintelligence and this time, it's not holding back. CEO Mark Zuckerberg has reignited the term Zuck Bucks; once a nickname for campaign donations, it's now shorthand for eight- and nine-figure signing packages aimed at luring the best AI minds in the world, according to a Reuters report. Warning! GuruFocus has detected 6 Warning Sign with META. Faced with talent losses and a disappointing release of its Llama 4 model, Meta has ramped up hiring and it's not being subtle. From a $14.3 billion investment in Scale AI to attempts at poaching Safe Superintelligence's Ilya Sutskever, Meta is making one thing clear: it wants back in the lead. Zuckerberg reportedly failed to recruit Sutskever but may be close to landing SSI co-founder Daniel Gross and NFDG's Nat Friedman. These aren't just star names they're magnets; and Meta hopes they'll help rebuild a team that's been bleeding to labs like OpenAI, Anthropic, and Google DeepMind. Meanwhile, Meta is forming an elite Superintelligence unit to push the boundaries of what AI can do. But internal divisions remain; Meta's Chief AI Scientist Yann LeCun has publicly questioned the long-term viability of large language models the very thing the company's top rivals are doubling down on. To complicate matters, Meta is betting on a mix of technologies: reasoning-based LLMs, multimodal AI, and even geopolitical hedges. For example, it's developing a new B40 chip tailored for the Chinese market, just in case export restrictions eventually ease. Zuckerberg's bet? Talent first; product later. This isn't classic M&A it's AI land-grabbing. Meta is willing to buy pre-product, pre-revenue startups if it means acquiring breakthrough IP and elite researchers. Profitability can wait; ASI supremacy can't. This article first appeared on GuruFocus. Sign in to access your portfolio
Yahoo
04-07-2025
- Business
- Yahoo
Zuck Bucks Are Back -- And This Time They're Fueling Meta's AI Comeback
Meta (META, Financials) is diving headfirst into the race for Artificial Superintelligence and this time, it's not holding back. CEO Mark Zuckerberg has reignited the term Zuck Bucks; once a nickname for campaign donations, it's now shorthand for eight- and nine-figure signing packages aimed at luring the best AI minds in the world, according to a Reuters report. Warning! GuruFocus has detected 6 Warning Sign with META. Faced with talent losses and a disappointing release of its Llama 4 model, Meta has ramped up hiring and it's not being subtle. From a $14.3 billion investment in Scale AI to attempts at poaching Safe Superintelligence's Ilya Sutskever, Meta is making one thing clear: it wants back in the lead. Zuckerberg reportedly failed to recruit Sutskever but may be close to landing SSI co-founder Daniel Gross and NFDG's Nat Friedman. These aren't just star names they're magnets; and Meta hopes they'll help rebuild a team that's been bleeding to labs like OpenAI, Anthropic, and Google DeepMind. Meanwhile, Meta is forming an elite Superintelligence unit to push the boundaries of what AI can do. But internal divisions remain; Meta's Chief AI Scientist Yann LeCun has publicly questioned the long-term viability of large language models the very thing the company's top rivals are doubling down on. To complicate matters, Meta is betting on a mix of technologies: reasoning-based LLMs, multimodal AI, and even geopolitical hedges. For example, it's developing a new B40 chip tailored for the Chinese market, just in case export restrictions eventually ease. Zuckerberg's bet? Talent first; product later. This isn't classic M&A it's AI land-grabbing. Meta is willing to buy pre-product, pre-revenue startups if it means acquiring breakthrough IP and elite researchers. Profitability can wait; ASI supremacy can't. This article first appeared on GuruFocus. Sign in to access your portfolio
