# Пов'язані статті щодо Large Language Models

Центр новин HTX надає останні статті та поглиблений аналіз на тему "Large Language Models", що охоплює ринкові тренди, оновлення проєктів, технологічні розробки та регуляторну політику в криптоіндустрії.

Shing-Tung Yau Invites Wang Hong and Deng Yu to Teach Back Home, After 20 Years of Rivalry, Tsinghua and Peking University Realize the Opponent Isn't Each Other

On July 23, the Fields Medal, the highest global mathematics award, was awarded to two Chinese mathematicians, Wang Hong and Deng Yu, both undergraduate classmates from Peking University's class of 2007. This marks the first time mathematicians who completed their undergraduate studies in mainland China have won the prize, with two laureates in the same year. Following the announcement, renowned mathematician Shing-Tung Yau publicly invited Wang and Deng to return to China to teach at Tsinghua University's Qiuzhen College. This invitation is the latest episode in a decades-long rivalry between Tsinghua and Peking University (PKU) for supremacy in Chinese mathematics. The competition traces back to a 1952 national university restructuring that transferred Tsinghua's entire mathematics department to PKU, leaving Tsinghua without a math program for nearly 30 years until its reinstatement in 1979. The rivalry intensified in 2005 with a public falling-out between Yau and his former student, PKU professor Gang Tian. Subsequently, Yau shifted his efforts to building Tsinghua's mathematics program from the ground up. He established the Yau Mathematical Sciences Center in 2009 and founded the Qiuzhen College in 2021, implementing an elite, accelerated "Math Leadership Program" to recruit prodigies as young as middle school students. The competition spans multiple fronts: recruiting top junior high and high school students from math olympiads, luring renowned scholars from overseas, and securing research funding. A new and fierce battleground has emerged in AI and mathematical reasoning. Both universities are heavily investing in AI research to enhance large language models' mathematical capabilities, with Tsinghua collaborating on projects like the STARE algorithm and PKU developing specialized models for geometry. Despite the intense competition, this rivalry has arguably elevated China's overall standing in mathematics. Chinese mathematicians now frequently publish in top journals and win major international awards. At the recent 17th Yau College Students Mathematics Contests, PKU students swept the top individual awards. Yau, presenting the awards, emphasized the need for inter-university collaboration to advance Chinese mathematics. The landscape is now shifting with the rise of AI in mathematical research. As AI begins to generate proofs at an unprecedented rate, the traditional metrics of competition between institutions may become less relevant. The future for mathematicians, as noted by Terence Tao, is navigating an era of "proof surplus" driven by AI, presenting a new common challenge that transcends university rivalries.

marsbit07/24 07:28

Shing-Tung Yau Invites Wang Hong and Deng Yu to Teach Back Home, After 20 Years of Rivalry, Tsinghua and Peking University Realize the Opponent Isn't Each Other

marsbit07/24 07:28

WEEX Labs Weekly Observation: The 'Power Restructuring' of AI Infrastructure and the 'Deep Dive Movement' into the Real Economy

WEEX Labs Weekly Review: AI Infrastructure's "Power Restructuring" and the "Deep Dive" into the Real Economy Mid-July 2026 marks a pivotal shift in the global AI industry. The allocation of computing power is transferring from cloud giants to compute resource owners, while the core value of AI is solidifying around its penetration into physical industry, moving beyond the race for model parameters. The era of fragmented model development is over, replaced by a capital-intensive, integrated chain driven by hard tech. Key developments this week include Meta's planned entry into the cloud computing market with "MetaCompute." This move by social media giants with massive GPU clusters challenges traditional cloud providers like AWS, integrating compute, models, and data into one-stop services, which will squeeze smaller rental providers and shift enterprise focus towards underlying model ecosystems. Chinese foundational models like DeepSeek-V4 and Tencent's Hy-3 are pushing towards "utility" status through open-source releases and extreme cost reductions via MoE architectures. This lowers entry barriers for enterprises, allowing them to focus resources on private deployment and deep business integration. Embodied intelligence, particularly humanoid robots, is transitioning from lab demos to real-world factory applications, driven by policies promoting large-scale, practical deployment in logistics and manufacturing. The value focus is shifting from spectacle to stable industrial data and real operational efficiency. Global governance, through forums like WAIC, is evolving from theoretical ethics to practical operational frameworks for "Sovereign AI," raising geopolitical compliance barriers and making auditability and data sovereignty core design requirements from the outset. WEEX Labs Insights: The current transformation shows AI's prosperity is deeply embedding into the fabric of global manufacturing. Strategic recommendations include: 1) leveraging open-source models for private, proprietary knowledge bases; 2) maintaining cloud provider diversity to avoid vendor lock-in from integrated model ecosystems; and 3) seeking opportunities in the "embodied infrastructure" supporting robots, such as data collection, industrial simulation, and factory AI adaptation services.

marsbit07/19 05:15

WEEX Labs Weekly Observation: The 'Power Restructuring' of AI Infrastructure and the 'Deep Dive Movement' into the Real Economy

marsbit07/19 05:15

Scaling Law a One-Size-Fits-All Solution? First Crystal Structure Manipulation Benchmark Shows Top Large Models Falling Short

Scaling Law Hits a Wall: New Benchmark Reveals AI's Struggles with Atomic-Level Material Manipulation A new benchmark called AtomWorld, developed by researchers, reveals a significant limitation in current large language models (LLMs). While powerful at understanding textual scientific knowledge, they perform poorly when tasked with physically manipulating atomic structures based on natural language instructions. The benchmark tests core atomic operations like replacing atoms, rotating structures, and expanding supercells. Results show that simply scaling up model size (Scaling Law) yields only modest and unstable improvements, particularly for tasks requiring strong 3D spatial reasoning and geometric planning. For instance, complex tasks like "rotating around a specific atom" see very low success rates even in top models like Claude Opus. This highlights a critical gap: textual knowledge does not automatically translate to reliable action in a physically constrained 3D space. The study argues that for AI in Science to progress, the focus must shift from just scaling language data (Language Scaling) to also scaling actionable capabilities (Action Scaling). This involves building training loops around "action-feedback-correction" cycles within simulated or real scientific environments. Ultimately, AtomWorld underscores that to become true lab assistants, AI models need to evolve beyond explaining knowledge to reliably executing precise, verifiable scientific actions.

marsbit07/15 03:56

Scaling Law a One-Size-Fits-All Solution? First Crystal Structure Manipulation Benchmark Shows Top Large Models Falling Short

marsbit07/15 03:56

US Large Models Moving Towards Closure, In the Name of Security

Summary: The U.S. government has intervened to regulate the release of advanced AI models, citing national security concerns. This analysis details the events surrounding the controlled rollout of Anthropic's Mythos 5 and Fable 5 models and OpenAI's restricted GPT-5.6 release, following a Trump administration executive order. It questions the technical basis for the government's actions, noting that internal company safety assessments did not deem the models an unacceptable risk. The article draws parallels to the 1990s "Crypto Wars," where similar attempts to control encryption technology ultimately failed, harming U.S. competitiveness. Experts warn that opaque, non-technical regulatory processes could stifle innovation, concentrate power, and undermine the economic logic of the AI industry, especially as Chinese models embrace open-source development. The piece concludes that this marks a potential turning point toward a pre-approval era for U.S. AI, creating uncertainty for developers and potentially slowing the broad societal diffusion essential for a technology's success.

链捕手06/27 15:21

US Large Models Moving Towards Closure, In the Name of Security

链捕手06/27 15:21

The World Cup has only been played for a few days, but some AI prediction models have already been crowned as oracles, while others have stumbled badly.

The 2026 FIFA World Cup has sparked significant interest not only on the pitch but also in AI-driven match prediction. Major models like Qwen, Copilot, and ChatGPT are being used to forecast outcomes, scores, upsets, red cards, and key player performances. Qwen gained early attention by accurately predicting Mexico's 2-0 win over South Africa (including a red card risk) and South Korea's 2-1 victory over the Czech Republic in the opening matches. Copilot's pre-tournament predictions had notable successes, such as correctly calling the Mexico 2-0 scoreline, South Korea's 2-1 win, and Brazil's 1-1 draw with Morocco. However, it also had clear misses, failing to predict upsets like Australia's 2-0 win over Turkey or Switzerland's draw with Qatar. ChatGPT provided detailed analytical reasoning, correctly predicting Mexico's 2-0 win, but its full-tournament predictions tended to favor favorites, missing several underdog results and draws. Tests pitting multiple models (ChatGPT, Gemini, Grok, Claude) against the same match, like Mexico vs. South Africa, showed varying predictions, with only some hitting the exact score. In summary, while AI models like Qwen have shown promising early results in specific match details, and others have had isolated successes, they collectively struggle to consistently identify upsets and underdog performances. AI is becoming an additional reference tool for prediction markets but is far from a definitive source.

marsbit06/16 03:53

The World Cup has only been played for a few days, but some AI prediction models have already been crowned as oracles, while others have stumbled badly.