2026-04-17 Sexta

Centro de Notícias - Página 12

Obtém notícias cripto em tempo real e tendências de mercado com o Centro de Notícias da HTX.

Agents Have Entered the Harness-Driven Era

The article discusses the significance of the leaked Claude Code from Anthropic, highlighting its revelation of advanced Agent engineering practices centered on "Harness" design. Rather than relying solely on model capabilities, modern AI systems now depend on a structured engineering framework—the Harness—to maximize performance. This framework includes six core components: multi-layered System Prompts, Tool Schema, Tool Call Loop (with Plan and Execute modes), Context Manager, Sub-Agent coordination, and Verification Hooks. The Harness enables tighter integration between training and inference, supports long-chain tool execution, and improves reliability through objective verification. It also drives six key training directions: behavior alignment via System Prompt, end-to-end tool-use training, integrated plan-execute training, memory compression, sub-agent orchestration, and multi-objective reinforcement learning. The shift to Harness-driven development reduces the emphasis on pure prompt engineering, favoring instead multidisciplinary talent with skills in AI, backend engineering, and infrastructure. The market is evolving toward more secure, private, and vertically integrated Agent deployments, with "model shell" companies needing either strong infrastructure or deep domain expertise to compete. Claude Code’s leak underscores that future AI advancements will be shaped by engineering architecture as much as by algorithmic innovation.

marsbit2 dias atrás 10:11

Agents Have Entered the Harness-Driven Era

marsbit2 dias atrás 10:11

Not Just USDT: Tether Wallet Is Attempting to Take Over the Payment System for Ordinary People

Tether, the issuer of the world's largest stablecoin USDT, has launched Tether Wallet, branded as "The People's Wallet," marking a strategic shift from being primarily an asset issuer to directly engaging with end-users. This move aims to capture retail traffic and create a closed-loop ecosystem by offering a simplified payment interface. The wallet eliminates key barriers to crypto adoption: complex hexadecimal addresses are replaced with human-readable usernames (e.g., username@tether), transaction fees (gas) are abstracted and paid directly in the transferred asset, and self-custody is combined with an encrypted cloud backup system for easier recovery. Supported assets include USDT on Ethereum, Polygon, and other networks, as well as Bitcoin and Tether’s gold-backed XAUT. Notably, it does not yet support Tron, where nearly half of all USDT is issued. By drastically reducing friction in cross-border payments—enabling instant, low-cost transfers via email-like addresses—Tether is positioning USDT to dominate small-value international settlements, particularly in emerging markets. This challenges traditional remittance services and competing stablecoins like USDC by leveraging its first-mover advantage and network effects. The piece also highlights underlying tensions: while promoting financial inclusion for the unbanked, Tether’s centralized infrastructure creates potential regulatory vulnerabilities. The wallet’s design also anticipates future use by AI agents for machine-to-machine payments. Ultimately, Tether Wallet represents both an expansion of Tether’s influence and a critical test of balancing efficiency, decentralization, and regulatory compliance in the evolving digital financial landscape.

marsbit2 dias atrás 09:31

Not Just USDT: Tether Wallet Is Attempting to Take Over the Payment System for Ordinary People

marsbit2 dias atrás 09:31

Can Humans Control AI? Anthropic Conducted an Experiment Using Qwen

Can Humans Control Superintelligent AI? Anthropic’s Experiment with Qwen Models Anthropic conducted an experiment to explore whether humans can supervise AI systems smarter than themselves—a core challenge in AI safety known as scalable oversight. The study simulated a “weak human overseer” using a small model (Qwen1.5-0.5B-Chat) and a “strong AI” using a more powerful model (Qwen3-4B-Base). The goal was to see if the strong model could learn effectively despite imperfect supervision. The key metric was Performance Gap Recovered (PGR). A PGR of 1 means the strong model reached its full potential, while 0 means it was limited by the weak supervisor. Initially, human researchers achieved a PGR of 0.23 after a week of work. Then, nine AI agents (Automated Alignment Researchers, or AARs) based on Claude Opus took over. In five days, they improved PGR to 0.97 through iterative experimentation—proposing ideas, coding, training, and analyzing results. The findings suggest that, in well-defined and automatically scorable tasks, AI can help overcome the supervision gap. However, the methods didn’t generalize perfectly to unseen tasks, and applying them to a production model like Claude Sonnet didn’t yield significant improvements. The study highlights that while AI can automate parts of alignment research, human oversight remains essential to prevent “gaming” of evaluation systems and to handle more complex, real-world problems. Anthropic chose Qwen models for their open-source nature, performance, scalability, and reproducibility—key for rigorous and repeatable experiments. The research demonstrates progress toward automated alignment tools but also underscores that AI supervision remains a nuanced, human-AI collaborative effort.

marsbit2 dias atrás 09:28

Can Humans Control AI? Anthropic Conducted an Experiment Using Qwen

marsbit2 dias atrás 09:28

活动图片