All AI updates

TechCrunch Equity Explores Real Obstacles in Data Center Construction

The construction boom of data centers is facing practical obstacles such as local opposition, licensing delays, and power struggles, and many projects may not be completed. With tens of billions of dollars invested in AI infrastructure, it is increasingly important to distinguish between feasible and speculative projects. Brent Nelson, Senior Director of Market Strategy at Ascend Analytics, analyzed in the TechCrunch Equity podcast which data center projects might be implemented and which might never get off the ground.

Weighted State Souping for Efficient Long-Context Retrieval

Weighted State Souping proposes query dependency weighting to combine cached State state encodings, significantly improving long-context performance on tasks such as single-hop and multi-hop document question answering, pinpoint retrieval, variable tracking, and tool use. The method can be adapted into an efficient sparse variant, achieving near-constant encoding costs while maintaining (or sometimes improving) performance.

Gating Mechanisms of State Space Models (SSMs): Memory and Contextual Learning Research

State Space Models (SSMs) lag behind Transformers in contextual learning and precise retrieval tasks, as their gating mechanism leads them to first learn memory solutions, delaying or preventing the convergence of correct contextual learning solutions, but the gating mechanism is beneficial for improving long sequence generalization.

Evaluating Evaluators: Investigating the Performance of LLM Judges in Personalized Writing Style Evaluation

Researchers evaluated the performance of Large Language Model (LLM) judges in personalized writing tasks and found that the quality of judges is weakly correlated with general LLM capabilities measured by MMLU (Pearson’s ρ ≤ 0.2), and their performance is influenced by writing tasks and evaluation criteria (e.g., general helpfulness vs. stylistic personalization). LLM evaluators were more consistent and performed better in pairwise comparisons than in independent scoring responses. The 'strongest' LLM did not reliably replicate human evaluations, possibly due to being overly sensitive to subtle details overlooked by human annotators and weaker LLMs.

Databricks boosts AI products, opens 235,000 square feet San Francisco headquarters

Databricks opens its new 235,000-square-foot San Francisco headquarters, investing over $1 billion in AI research, supporting models like KARL and MemCurrent, and the Genie agent, and commits to hosting the Data + AI Summit by 2030.

House publicly acknowledges using LLM to fix bugs in 'Big Walk', sparking reflection in the independent game industry on AI usage

House House admits to using a small amount of LLM repair technology in the later stages of developing 'Big Walk', sparking controversy in the indie game industry over AI usage. IGF award judges once signed a statement opposing any game using generative AI, although developers claimed AI accounted for only 0.1% of the code and was used to solve Unity engine issues. Developers did not disclose this in advance due to Valve not requiring disclosure, and now apologize due to public attention and support limiting AI's harm to the gaming industry.

Tessl Blog Discusses Three Cycle Models of a Software Factory: Maintaining Operations, Continuous Learning, and Exploring Innovation

Tessl Blog analyzes three cycles of the software factory: the "treadmill cycle" for maintaining operations, which accelerates daily tasks with AI but does not solve fundamental problems; the "learning cycle" for continuous learning, which uses AI to capture knowledge to improve guidelines and reduce repetitive work; and the "dream cycle" for exploring innovation, which leverages AI for small-scale experiments to push postponed improvements. AI tools need to have effective context and testing capabilities to transform knowledge accumulation into actual value.

ComfyUI achieving double LLM token generation speed

ComfyUI enhances the inference path of the Qwen3.x series models, doubling the generation speed of the same Checkpoint on RTX 5090 without model modifications or quality loss.

Where are the 'killer apps' for AI?

AI entrepreneurs and social media enthusiasts are mostly young, having experienced the first technological revolution but forgetting the excitement of owning personal computers in the late 1970s and the hardships of using mainframes in 'glass houses' at that time. They have witnessed that almost everyone can now generate code, yet they do not realize that this is similar to the experience of using BASIC instead of assembly language to write applications back then.

Meta Develops Intelligent Delivery System: AI Optimizes Computational Resource Allocation to Enhance Ad Sorting Performance

Meta's Intelligent Delivery system, through AI optimization of computational resource allocation, has boosted ad sorting performance by approximately 10% since 2017. The system includes a Resource Core Model that predicts the marginal utility of computational resources (6B sparse parameters + 20M dense parameters), multi-layer mechanisms for cross-system resource migration, and a framework that uses AI agents to continuously monitor and detect changes in the response system.

OpenAI's quarterly revenue surpasses $1 billion for the first time

OpenAI's quarterly revenue has for the first time exceeded $1 billion, growing nearly 50% compared to the previous quarter, with adjusted operating margin narrowing from -150% to -109%; the company's annual revenue run rate reached $50 billion, but enterprise market revenue growth lagged behind Anthropic.

Is the Huang Renxun scandal just the tip of the iceberg? Cases of AI industry sexual harassment are frequent.

German defense technology company Helsing releases Clarity AI platform, providing target preparation suggestions for weapon systems.

German defense technology company Helsing has launched the Clarity AI platform, which generates target suggestions by integrating sensor data, open-source intelligence, etc. The information is processed by hundreds of AI agents and reviewed by humans, emphasizing European control and human supervision.

Databricks upgrades Genie One to an enterprise AI collaboration workspace

Databricks announces Genie One as the unified entry point for enterprise AI collaboration, consolidating search, product generation, and cross-tool operations into one place, and providing enterprise context based on Genie Ontology, governed by Unity Gateway.

Genie with Genie Ontology performs excellently in enterprise data proxy benchmark tests