AI Technology Daily: Key AI Industry Updates March 2, 2026

1. Palantir stock jumps as Anthropic AI aids US airstrike on Iran

Palantir shares rose nearly 4% to $142.6 pre‑market after reports that its collaboration with Anthropic supplied AI technology that improved the precision of a US airstrike on Iran, marking a notable military application of generative AI.

2. OpenAI’s DoD agreement draws criticism

OpenAI signed a contract with the U.S. Department of Defense, with financial terms undisclosed, aimed at ensuring its models are not used for mass surveillance or autonomous weapons. The deal sparked industry concerns about AI ethics and governance, potentially influencing future AI development directions.

3. Alec Radford’s new work promises significant model performance gains

Alec Radford announced a new optimization solution for large models, backed by tens of millions of dollars in financing. The approach reportedly delivers notable performance improvements, positioning it for substantial market share growth and suggesting a possible shift in AI industry dynamics.

4. OpenAI scales single‑master PostgreSQL for massive query loads

OpenAI successfully extended a single‑master PostgreSQL deployment to support the ChatGPT service and its API platform at millions of queries per second. By optimizing query processing, PostgreSQL performance reached new heights, offering strong support for large‑scale data handling.

5. Huawei team wins CVRPLib BKS global challenge with innovative algorithm

The Huawei Hong Kong team captured first place in the CVRPLib BKS global competition, employing a “large‑model + evolutionary computation” technique that restructured the design paradigm. Although financing details were not disclosed, the algorithm’s performance outpaced industry peers, promising a sizable market impact in operations research.

6. Alibaba’s REG4Rec pushes recommendation system capabilities

Alibaba introduced REG4Rec, a recommendation system backed by over $1 billion in financing. The platform achieves millisecond‑level feature understanding and ranking decisions, enabling efficient matching of content and products, and signaling a new stage for personalized recommendation technology.

7. Notion releases its first open‑source weight model, MiniMax M2.5

Notion unveiled MiniMax M2.5, the first open‑source weight model integrated into a mainstream productivity tool, indicating that domestically developed large models are entering core software ecosystems. Financing details were not disclosed, but the move is expected to increase market share and break the dominance of closed‑source models.

8. Peking University publishes generative CL‑Bench, advancing GENIUS model

The Peking University team released the CL‑Bench benchmark for generative models, showcasing significant performance gains of the GENIUS model, which unifies generation and understanding while centering on data‑driven AI. Funding exceeded $10 million, and the innovation points to new directions for AI research.

9. OpenAI may skip GPT‑5.3 and launch GPT‑5.4 directly

Internal GitHub pull‑request activity suggests OpenAI could bypass a GPT‑5.3 release and move straight to GPT‑5.4. Performance metrics and market impact remain unspecified, but the leak hints at an accelerated innovation pace in the AI field.

10. Peking University and OceanBase tackle long‑video retrieval with LoVR benchmark

Collaborating with OceanBase, Peking University introduced the LoVR benchmark to fill a gap in long‑video retrieval evaluation. The technology addresses three major challenges in the domain, delivering superior performance and promising to raise industry standards.

11. Alibaba’s Qianwen 3.5 dominates Hugging Face leaderboard

Alibaba Cloud’s Qianwen 3.5 secured the top four positions on Hugging Face’s ranking, demonstrating strong technical capability. While financing details were not disclosed, the open‑source strategy continues to expand its influence, underscoring AI’s deep integration into everyday life.

12. Shanghai Jiao‑Tong University proposes JTok module to cut compute demand

Shanghai Jiao‑Tong University unveiled the JTok module, which reduces computational requirements to one‑tenth of typical large‑model baselines. The module has attracted multi‑million‑dollar financing and is poised to accelerate large‑model development.