Artificial intelligence is experiencing a pivotal moment in 2025, as organizations worldwide grapple with the challenge of moving from experimentation to enterprise-wide adoption. At the same time, leading voices in the field are questioning the prevailing approaches that have dominated the AI landscape in recent years. The departure of Yann LeCun, a Turing Award winner and one of the fathers of modern AI, from his post as Meta’s chief AI scientist, marks a turning point that reflects both the growing pains and the promise of the next era of machine intelligence.
LeCun’s exit, announced last week, comes as Meta pivots its research focus under new leadership. According to reporting from Big Technology and confirmed by LeCun himself on LinkedIn, he will leave Meta by the end of 2025 to launch a new startup dedicated to what he calls the Advanced Machine Intelligence research program. This move is more than a career change—it signals a philosophical shift in how AI’s future could unfold.
In a world captivated by the capabilities of large language models (LLMs), LeCun has long sounded a note of caution. On the Big Technology podcast in May, he declared, “We are not going to get to human-level AI just by scaling LLMs.” He described LLMs as “systems with gigantic memory and retrieval ability, not a system that can invent solutions to new problems.” For LeCun, the path to true artificial intelligence lies elsewhere.
LeCun’s critique is rooted in his decades-long advocacy for self-supervised learning—a method where AI systems learn from data without explicit human labeling. He famously quipped at the AAAI conference in 2020, “If artificial intelligence is a cake, self-supervised learning is the bulk of the cake. The next revolution in AI will not be supervised, nor purely reinforced.” While LLMs do use a form of self-supervised learning, LeCun argues they are fundamentally limited because they only digest human-generated text and lack the ability to learn from the unpredictability of the real world.
“They do not learn how the world works from sensory data like children do by observing and interacting with the world,” LeCun has explained. He believes that current models fail to internalize basic concepts like gravity, depth, or object permanence, and therefore cannot predict consequences and counterfactuals with the flexibility of a human mind. This, he contends, is why LLMs require enormous training data for even basic tasks and still struggle to generalize beyond what they’ve seen.
LeCun’s next venture will focus on building AI systems that “understand the physical world, have persistent memory, can reason, and can plan complex action sequences,” as he stated in his LinkedIn announcement. The goal is nothing less than to spark “the next big revolution in AI.” The industry, he suggests, has mostly ignored these ambitions in its rush to scale up existing models.
Meanwhile, the business world is wrestling with its own set of AI challenges. According to the 2025 IMD AI Maturity Index, which analyzed data from the world’s 300 largest companies, the race to scale AI after a year of experimentation in 2024 has proven tougher than many expected. As reported by IMD researchers and published by TONOMUS, Gartner predicted that 30% of AI initiatives would be abandoned after the proof-of-concept stage by the end of 2025. The reasons are familiar but stubborn: legacy IT systems, siloed data, skills shortages, and increasing regulatory scrutiny.
Yet, the companies that overcome these hurdles are reaping tangible rewards. The IMD Index found that the top 100 firms investing across five key dimensions of AI maturity—leadership, people, technology, governance, and purpose—achieved average year-over-year revenue growth in 2025. These leaders are not just deploying the latest models but are embedding AI into the very fabric of their organizations.
Take the automotive sector, for example. Volkswagen Group and Mercedes-Benz Group are redefining mobility with software-defined vehicles. Volkswagen’s in-house AI copilots personalize driving behavior, while Mercedes-Benz’s MB.OS™ platform delivers continuous vehicle performance optimization via over-the-air updates. In China, BYD partners with PTC to unify design and production data in a single AI environment, accelerating innovation and scale.
Manufacturing giants like Siemens and GE Aerospace are embedding AI into the entire design-to-production cycle. Siemens’ collaboration with Nvidia enables real-time simulation and predictive maintenance on the factory floor, while GE Aerospace uses AI for parts forecasting and quality assurance. These efforts show how AI, when deeply integrated, transforms products and operations from the inside out.
The financial sector, too, is embracing AI at scale. Mastercard’s generative AI models perform real-time fraud detection across millions of transactions daily. KKR blends AI with human expertise for investment modeling, and Ping An Insurance’s in-house research arm automates underwriting and claims. Meanwhile, Goldman Sachs and Visa have formalized Responsible AI oversight, ensuring transparency and accountability.
Retailers are not far behind. Walmart’s Wallaby™ LLM assists associates and optimizes merchandising, while Kroger deploys predictive analytics and smart-shelf technology to anticipate demand and reduce waste. Unilever’s Beauty AI Studio™ crafts personalized product formulations and marketing, with L’Oréal and Nike using AI for brand storytelling and design. These companies invest in employee AI training to ensure adoption permeates every level of the business.
In energy and utilities, Equinor and Engie leverage AI for grid forecasting and carbon tracking, while SLB and PTT use digital twins for real-time analytics that reduce both costs and environmental impact. Healthcare leaders like Medtronic and CVS Health integrate AI into diagnostics and patient care, and pharmaceutical giants AstraZeneca and Merck use LLMs to accelerate drug discovery. Sanofi’s partnership with OpenAI signals a new era of AI-enabled R&D.
Telecom companies like Deutsche Telekom and KDDI use AI to predict network outages and optimize bandwidth. Not surprisingly, technology titans Nvidia, Microsoft, and Alphabet remain at the forefront, leading global AI infrastructure and foundation model development. Their platforms, such as Nvidia’s AI chips, Microsoft’s Copilot® suite, and Alphabet’s Gemini models, are setting new benchmarks for the industry.
Yet, even as AI becomes ever more embedded in business, critics like Professor Amar Bhidé warn against overhyping LLMs, arguing they “flatter rather than enlighten” and that executives must distinguish between calculable risk and true uncertainty. The IMD research underscores that successful scaling of AI is as much about organizational change, governance, and workforce upskilling as it is about technology itself.
As LeCun’s bold new direction and the struggles of corporate AI adoption both show, the future of artificial intelligence will be shaped by those who not only trust and govern the technology, but who are willing to rethink its very foundations. The next chapter in AI will require systems that can learn, reason, and adapt like humans—and organizations ready to transform from the inside out.
