# Artikel Terkait Models

Pusat Berita HTX menyediakan artikel terbaru dan analisis mendalam mengenai "Models", mencakup tren pasar, pembaruan proyek, perkembangan teknologi, dan kebijakan regulasi di industri kripto.

Zhejiang University Research Team Proposes New Approach: Teaching AI How the Human Brain Understands the World

A research team from Zhejiang University published a paper in *Nature Communications* challenging the prevailing notion that larger AI models inherently think more like humans. They found that while model performance on recognizing concrete concepts improved as parameters increased (from 74.94% to 85.87%), performance on abstract concept tasks slightly declined (from 54.37% to 52.82%) in models like SimCLR, CLIP, and DINOv2. The key difference lies in how concepts are organized. Humans naturally form hierarchical categories (e.g., grouping a swan and an owl into "birds"), enabling them to apply past knowledge to new situations. Models, however, rely heavily on statistical patterns in data and struggle to form stable, abstract categories. The team proposed a novel solution: using human brain signals (recorded when viewing images) to supervise and guide the model's internal organization of concepts. This method, termed transferring "human conceptual structures," helped the model learn a brain-like categorical system. In experiments, the model showed improved few-shot learning and generalization, with a 20.5% average improvement on a task requiring abstract categorization like distinguishing living vs. non-living things, even outperforming much larger models. This research shifts the focus from simply scaling model size ("bigger is better") to designing smarter internal structures ("structured is smarter"). It highlights a new pathway for developing AI that possesses more human-like abstract reasoning and adaptive learning capabilities.

marsbit04/05 04:41

Zhejiang University Research Team Proposes New Approach: Teaching AI How the Human Brain Understands the World

marsbit04/05 04:41

Cursor vs. Anthropic and OpenAI: Thanks for Raising Me, Now I'm Here to Take the Market

Cursor, a VS Code plugin initially built on OpenAI's API, has transitioned from a dependent customer to a formidable competitor by launching its proprietary coding model, Composer 2. This model reportedly outperforms Claude Opus 4.6 on key benchmarks at one-tenth the cost. The case exemplifies a critical strategic dilemma in tech—when to open or close an API. The authors propose a framework: opening an API risks eroding a company’s moat if competitors can use it to bootstrap their own products and aggregate demand, eventually enabling vertical integration. This is especially risky in AI, where API outputs can directly improve a rival’s model training and product refinement—exactly what Cursor achieved by leveraging OpenAI and Anthropic models to gather user data and refine its own offering. Companies then face two choices: restrict API access (like Twitter, which closed its API to protect its social graph) or keep it open but find alternative moat, such as network effects or Lindy effects (like crypto protocols, e.g., Morpho). The authors predict that leading AI companies (like OpenAI and Anthropic) will likely restrict access to their most advanced models over time, as switching costs remain low, network effects are weak, and distillation techniques reduce training costs. This could stifle consumer AI innovation but create opportunities for open alternatives.

marsbit03/31 07:35

Cursor vs. Anthropic and OpenAI: Thanks for Raising Me, Now I'm Here to Take the Market

marsbit03/31 07:35

From Power to Chips: How Ordinary People Can Participate in the Wealth Opportunities of the AI Era

From Power to Chips: How Ordinary People Can Participate in the Wealth Opportunities of the AI Era This article analyzes the AI industry through a five-layer "AI stack" framework: energy, chips, cloud infrastructure, models, and applications. It argues that while public attention focuses on the top application layer (e.g., ChatGPT), the vast majority of capital investment and profits are currently concentrated in the underlying infrastructure layers. Key points include: - An estimated $700 billion in annual capital expenditure is flowing into AI infrastructure (energy, chips, data centers), not applications. - Infrastructure companies (Nvidia, TSMC, ASML) show massive profits and near-monopolies, while model companies (OpenAI, Anthropic) experience rapid revenue growth but burn enormous cash due to compute costs. - Historical parallels are drawn to the electricity revolution and internet infrastructure boom, where infrastructure builders captured most early value. - The article advises investors to focus on infrastructure layers currently generating concentrated profits, while acknowledging future value may shift to applications as the market matures. - Risks include capital misallocation, supply chain concentration, and efficiency breakthroughs (like DeepSeek's lower-cost models) that could disrupt current assumptions. The conclusion emphasizes understanding this layered structure, tracking capital flow, and participating at appropriate levels based on risk tolerance and expertise.

marsbit03/16 08:17

From Power to Chips: How Ordinary People Can Participate in the Wealth Opportunities of the AI Era

marsbit03/16 08:17

NVIDIA's Jensen Huang Latest Article: The 'Five-Layer Cake' of AI

NVIDIA's Jensen Huang articulates AI not merely as a software application but as a fundamental infrastructure, comparable to electricity or the internet, in a layered "five-layer cake" structure. This stack begins with **Energy** as the foundational constraint, powering real-time intelligence generation. Above it, **Chips** convert energy into computational power efficiently. The **Infrastructure** layer comprises data centers and systems that function as "AI factories." **Models** form the next layer, processing diverse data types like language, biology, and physics. At the top, **Applications**—such as drug discovery, autonomous vehicles, and robotics—create economic value. Huang emphasizes that AI is an industrial-scale transformation, driving massive global infrastructure expansion requiring trillions in investment and a skilled workforce—from electricians to network technicians—beyond just computer scientists. He notes that AI has recently crossed a threshold: models are now reliable enough for widespread use, reducing hallucinations and improving reasoning, which accelerates real-world applications. Open-source models, like DeepSeek-R1, further propel growth across the entire stack. This infrastructure revolution will reshape energy consumption, manufacturing, labor, and economic growth. Every company and country will participate, though the field remains early-stage, with vast opportunities and responsibilities ahead.

marsbit03/10 14:18

NVIDIA's Jensen Huang Latest Article: The 'Five-Layer Cake' of AI

marsbit03/10 14:18

Tokens, Models, and Bubbles: The Crypto × AI Game in the Primary Market

Based on a two-year retrospective, this article analyzes the convergence of Crypto and AI from a primary market perspective. Initially, the crypto space heavily promoted "Crypto Helps AI" through three main narratives: computation power tokenization, data tokenization, and model tokenization. However, these efforts largely resulted in what the author calls a "tokenization illusion"—projects that issued tokens but lacked real product-market fit or sustainable business models. The piece critiques these approaches: decentralized compute networks often fail to meet enterprise reliability standards; tokenized data struggles with supply-demand alignment due to low user motivation and high professional requirements; and model tokenization is fundamentally flawed since AI models are non-scarce, easily replicable, and depreciate quickly. Additionally, projects focusing on verifiable inference (like ZKML or OPML) are solutions in search of a problem, as real-world AI failures are rarely due to malicious tampering but rather design errors or misconfigurations. The author references Vitalik Buterin’s updated views, which now present a more balanced perspective compared to two years ago. Buterin outlines four quadrants of Crypto × AI integration: two where crypto (especially Ethereum) provides trustless, economic layers for AI agents and private interactions, and two where AI enhances crypto—through local LLMs acting as user shields for security and AI improving market efficiency and DAO governance. The conclusion emphasizes that meaningful progress lies at the intersection of both fields, beyond mere tokenization or speculative narratives, and expresses hope for more substantive developments in the future.

比推02/12 06:16

Tokens, Models, and Bubbles: The Crypto × AI Game in the Primary Market

比推02/12 06:16

活动图片