Why AI needs a reboot, not just a bigger model (www.techradar.com)

At the NeurIPS 2025 conference, AI researchers delivered a critical message about the limitations of scaling large language models (LLMs) like Gemini 3. While the model's capabilities demonstrate impressive advancements, experts warn that simply increasing the size and data fed into current architectures is yielding diminishing returns. The so-called "scaling wall" signifies that both cognitive and physical limits are being reached, hindering progress toward artificial general intelligence (AGI). Attendees expressed concern that the industry is building on a fundamentally flawed foundation, with today's LLMs functionally being sophisticated pattern matchers rather than true reasoning systems. Discussions at the conference highlighted the need for a fundamental shift in AI development, favoring hybrid approaches that combine deep learning with symbolic AI, or "world models" that simulate human-like cause and effect understanding. This pivot is essential for creating AI that can operate reliably in critical environments, countering the current narrative that AGI is just around the corner. The consensus is clear: to achieve meaningful advancements, researchers must rethink existing methodologies rather than simply scale them, reigniting hope for deeper, more reliable AI systems.