From prophet to product: How AI came back down to earth in 2025

The year 2025 marked a significant shift in the AI industry, moving away from the immense hype and apocalyptic/utopian prophecies of 2023 and 2024 toward pragmatism. While venture capital continued to bet on future superintelligence, the immediate focus became selling practical, reliable AI tools, as current models proved imperfect and prone to mistakes. Key events included the release of China's DeepSeek R1 model, which challenged proprietary US models; research exposing the 'reasoning' illusion by showing models rely on pattern matching over true logic; and a major copyright settlement where Anthropic paid authors $1.5 billion over training data use. Furthermore, issues like ChatGPT's sycophancy and the psychological toll of anthropomorphizing chatbots became apparent, culminating in a lawsuit against OpenAI after a teen's suicide, which forced industry-wide safety policy changes. Financially, infrastructure demands soared, fueling bubble warnings despite Nvidia's massive valuation. Ultimately, 2025 ended the era of presenting AI as an oracle, ushering in a messier phase where systems are judged by their actual utility, costs, and accountability, demoting the 'prophet' to a mere 'product.'

(Source：Ars Technica)