In 2026, Richard Stallman remains a strong critic of the current AI boom. During his talks, he has described Large Language Models (LLMs) as “pretend intelligence,” arguing that they generate convincing text without genuine understanding or reasoning. His concern is that calling these systems “intelligent” can mislead people about what they actually do.
I agree with Stallman to some extent. Today’s LLMs are based on statistical learning rather than human-like understanding, and they should not be trusted blindly. They are powerful tools for generating text, code, and analysis, but they can also produce incorrect or fabricated information with great confidence.
Where I differ from Stallman is that I believe LLM technology itself has enormous value. The key question is not whether AI should exist, but whether it can be developed transparently and under users’ control. I would like to see more open-weight and open-source AI models, following the direction of projects such as Meta’s Llama, instead of a future dominated entirely by proprietary AI systems.
Today, several major LLM families dominate both consumer use and enterprise automation, including OpenAI’s GPT, Anthropic’s Claude, Google’s Gemini, Meta’s Llama, Alibaba’s Qwen, and DeepSeek. Each has different strengths, from software engineering and agentic workflows to multilingual reasoning and cost-efficient deployment.
Beyond the models themselves, new AI infrastructures are also emerging. Daniel Miessler’s AI Fabric is not a Large Language Model, but an architectural framework for orchestrating multiple AI models and tools into a unified system. This represents another important direction for AI development alongside increasingly capable foundation models.
The future of AI should combine openness, transparency, scientific scrutiny, and user freedom. Open models allow researchers, developers, and society to understand how these systems work, improve them collaboratively, and reduce dependence on a handful of proprietary vendors.