OpenAI officially introduced the GPT-5.2 model on its tenth anniversary, touting it as the most powerful for professional knowledge work. However, users report that daily conversations are bland, safety review is excessive, and it lacks emotional intelligence. It also scores lower than previous models like Claude Sonnet 3.7 on common sense reasoning tests such as SimpleBench. Concurrently, token consumption and costs have increased, with some functions like cross-application collaboration being strictly limited. Users find its programming capability and ASCII art creation neither visually nor practically outstanding; some have collectively criticized it for lacking empathy and having inflexible safety mechanisms.

GPT-5.2 achieved higher scores in some industry benchmark tests by drastically increasing token consumption, making it less efficient and more costly compared to models like Gemini 3.0 Pro. It failed to achieve superiority in some evaluations like LiveBench and MMMU-Pro. The community questions OpenAI's 'false marketing,' with real user experience revealing noticeable hallucination phenomena and a conservative conversational style. Some core team members have left the company, citing its trend toward commercialization and suppression of negative research.

In December 2025, ByteDance's Doubao large model launched a new AI phone assistant system service supporting complex cross-app operations with full-process privacy protection. It debuted on the engineering prototype of the Nubia M153, a collaborative effort with ZTE, selling out immediately upon launch. This drove ZTE's A-share market value to exceed 210 billion yuan. The related AI task processing speed leads industry competitors, attracting third-party restrictions and regulatory scrutiny. Currently, the Doubao large model already covers 300 million devices and cooperates with 80% of mainstream automakers, driving the formulation of industry standards for AI operation permissions.

U.S. President Trump signed an executive order aiming to unify regulatory standards for artificial intelligence at the federal level, restricting states from implementing different AI laws. It also establishes an AI litigation task force to challenge local laws. Tech companies generally welcome the new policy, but media outlets worry it may lead to relaxed regulation and associated risks.

Impacted by declining corporate earnings expectations and tight AI chip supply, AI-related tech stocks like Broadcom, Oracle, and NVIDIA saw significant declines, reigniting the 'bubble' debate in the AI sector. Oracle delayed the construction schedule for a data center being built for OpenAI until 2028. Infrastructure investments in the UK, US, and elsewhere face conflicts over labor and resource supply, with data centers consuming substantial societal construction resources.

Google opened access to Gemini Deep Research capabilities and a unified Interactions API, allowing developers to integrate automated knowledge extraction and structured report generation into enterprise-level application scenarios. Concurrently, the AI Agent standard (AAIF) and platforms like Microsoft launched a new generation of Agent development and management tools, promoting open standards and multi-model integrated collaboration.

AI automates repetitive, foundational tasks in areas like ad placement and media planning, marginalizing a large number of entry-level positions. The path to cultivating managerial and multi-skilled talents is now facing a risk of disruption. Major institutions are exploring ways to involve newcomers earlier in core data analysis and AI co-operation to bridge the 'experience gap'.

Disney announced a $1 billion cooperation with OpenAI, licensing over 200 of its classic characters to the Sora video platform, making it the first major IP content provider for Sora. Users can custom-generate related animation videos. The industry views this agreement as a milestone for comprehensive commercial collaboration between the content industry and large models.

Globally, data center investment, particularly in AI data centers, is rapidly expanding, with private sector annualized investment exceeding $41 billion, approaching the scale of government infrastructure like transportation. Labor shortages and limited energy supply have already increased national debt and prolonged social construction project cycles. Large AI projects consume vast amounts of electricity, land, and high-skilled labor, sparking widespread discussion on sustainability and fairness.

Researchers like Tim Dettmers argue that the cost-effectiveness of GPUs peaked after 2018, and further model scaling yields diminishing returns for increasing costs. Foundational deep learning algorithms are approaching their optimum. Future reliance on exponential investment is unsustainable, and the industry may need to shift towards 'reliability and multi-scenario integration' rather than endless parameter scaling.