Technology TrendsNews

Explores the latest innovations, protocol upgrades, cross-chain solutions, and security mechanisms in the blockchain space. It provides a developer-focused perspective to analyze emerging technological trends and potential breakthroughs.

"Agents' Last Exam", Claude Fable 5 Actually Loses to GPT 5.5

Surprisingly, in the newly released "Agents' Last Exam" (ALE) benchmark from UC Berkeley, GPT-5.5 has outperformed the recently launched and highly-regarded Claude Fable 5. ALE tests AI agents on their ability to perform real-world tasks across 55 professional domains—such as 3D modeling in Siemens NX, creating game scenes in Unreal Engine, and visual effects work in Adobe After Effects—by granting them full GUI and command-line access. In the core task completion rate ranking, GPT-5.5 configurations secured the top two spots (24.0% and 23.0%), while Claude Fable 5 with Claude Code came in third (22.0%). Notably, the highest pass rate was only 24%, and the most difficult "Last-Exam" tier saw most top models, including GPT-5.5 and Fable 5, scoring zero. The benchmark also revealed significant cost and efficiency gaps: Fable 5 spent over four times more money than GPT-5.5's most expensive configuration for a slightly lower score, and was much slower. ALE differs from previous knowledge-based benchmarks by evaluating practical "ability to do" rather than static knowledge retrieval. Its tasks are derived from real expert projects, automatically scored, and designed to prevent cheating through a rotating pool of private challenges. The results suggest that high performance on traditional benchmarks does not necessarily translate to proficiency in complex, open-ended real-world work. The study also notes that agents often fail by prematurely declaring tasks complete without proper verification, and that no single model excels uniformly across all diverse domains.

marsbit06/12 05:01

"Agents' Last Exam", Claude Fable 5 Actually Loses to GPT 5.5

marsbit06/12 05:01

The Recursive AI Anthropic Warned About: Tian Yuandong's New Company Has Just Taken the "First Step"

Anthropic recently highlighted the rapid progress toward "recursive self-improvement," where AI systems autonomously design and train their successors. In response, Recursive Superintelligence, a new company co-founded by former Meta researcher Tian Yuan Dong, has publicly demonstrated its first step toward automating AI research. The company released a system designed to autonomously execute the full AI research cycle: generating ideas, implementing code, running experiments, and learning from results. It validated this approach by achieving state-of-the-art results on three diverse benchmarks: 1. **NanoChat Autoresearch:** Optimizing a small language model's validation loss under a fixed 5-minute GPU budget, improving upon the community's best result. 2. **NanoGPT Speedrun:** Reducing the time to train a GPT model to a specific loss on 8 H100 GPUs from 79.7 seconds to 77.5 seconds, beating a highly optimized, human-driven community effort. 3. **SOL-ExecBench:** Improving the overall score on NVIDIA's suite of 235 GPU kernel optimization tasks by 18%, closing the gap to the hardware limit. The system discovered novel optimizations in this highly specialized domain without direct human expertise. Recursive's system operates as a general framework, capable of parallel exploration and cross-task knowledge transfer while incorporating safeguards against reward hacking. The company, backed by $650M in funding and a star-studded team including Richard Socher and Alexey Dosovitskiy, aims to create AI that recursively enhances its own research capabilities. This development represents an early but concrete move toward a new paradigm where AI accelerates its own advancement. It occurs alongside Anthropic's warnings about the need for industry coordination and potential pauses when recursive self-improvement thresholds are reached, highlighting the dual trajectory of rapid technical progress and growing calls for careful stewardship.

marsbit06/12 04:12

The Recursive AI Anthropic Warned About: Tian Yuandong's New Company Has Just Taken the "First Step"

marsbit06/12 04:12

Sequoia Dialogue with Jensen Huang: Computing Model Undergoes a 60-Year Transformation; You Won't Be Replaced by AI, But You Will Be Dimensionality-Reduced by 'Those Who Master AI'

NVIDIA founder and CEO Jensen Huang, in a conversation with Sequoia Capital's Konstantine Buhler, argues that we are witnessing the most significant computing shift in 60 years—from retrieval-based to generative computing. Instead of just storing and retrieving data, future systems will generate highly personalized content (text, images, video) on demand, powered by massive "AI factories." Huang envisions a global "intelligence network" that will envelop the planet, following the historical patterns of energy and communication grids. He outlines a five-layer investment framework: 1) Energy, 2) Chips/Computers, 3) Infrastructure (data centers), 4) AI Models, and 5) Applications. He predicts this ecosystem will reach a scale of $20 trillion annually. Crucially, Huang pushes back against fears of AI-driven job loss. He distinguishes between specific "tasks" (e.g., typing, analyzing images) and overall "jobs" (e.g., CEO, radiologist). While AI automates tasks, it increases efficiency and demand for the higher-value problem-solving aspects of professions, thus creating more jobs and "up-leveling" careers. The real risk, he asserts, is not being replaced by AI, but being outperformed by someone who effectively leverages it. He urges everyone to embrace AI as a tool for augmented capability and innovation.

marsbit06/12 02:59

Sequoia Dialogue with Jensen Huang: Computing Model Undergoes a 60-Year Transformation; You Won't Be Replaced by AI, But You Will Be Dimensionality-Reduced by 'Those Who Master AI'

marsbit06/12 02:59

"I Don't Need a Better Model Anymore": A Panorama of AI Users Under a Reddit Hot Post

Titled "I Don't Need a Better Model Anymore": AI User Reactions on Reddit Anthropic recently released Claude Fable 5, its first publicly available 'Mythos'-tier model, achieving 80.3% on the SWE-Bench Pro benchmark and significantly outperforming its predecessor and competitors. However, a viral Reddit post titled "Claude Fable made me realize I don't need better models anymore" highlighted a growing user sentiment of "good enough." Top comments expressed "model fatigue," with users stating that earlier models like Opus 4.5/4.8 already sufficed for their workflows. High cost was a key concern, as Fable 5's API is nearly twice the price of Opus 4.8, with users questioning the return on investment and suggesting the field has hit a plateau. The most frequent complaint targeted Fable 5's stringent safety filters. Designed to intercept high-risk requests (e.g., cybersecurity), the system was perceived as overly conservative. Users reported frequent rejections for routine security-related tasks, leading to automatic fallbacks to the older Opus model. Paying users were particularly frustrated, feeling they paid a premium for a less usable product. Dissenting voices came from users with heavy, complex tasks. For workloads like high-energy physics simulations with thousands of code lines, Fable 5's improved long-context understanding and error detection represented a significant, worthwhile leap—described as moving from a "college player to an NBA starter." The debate underscores a divergence between benchmark performance and practical utility. For most users, current models meet their needs, making further advances relevant only for extreme use-cases. The discussion also raised concerns about a potential "Public AI Freeze," where the most powerful models (like the restricted Mythos 5) remain exclusive to enterprises and governments, while public offerings stagnate. The launch presents two report cards: one of technical excellence and another of user skepticism. Fable 5's ultimate reception may depend on Anthropic's ability to refine its safety filters and justify its cost for specialized, high-demand users.

marsbit06/12 02:52

"I Don't Need a Better Model Anymore": A Panorama of AI Users Under a Reddit Hot Post

marsbit06/12 02:52

a16z Crypto Partner: Cash Flow is the Moat

Cash Flow as the Moat: A Playbook for Crypto Founders Historically, the most enduring businesses have been built by positioning themselves within the "flow of funds"—facilitating the creation and transfer of value in a network and extracting a portion of it. Cryptocurrency is the first modern technology natively built for this purpose. For startups, failing to architect products and businesses to leverage these principles means missing a major opportunity. Blockchains are inherently network businesses. Each transaction settles on a shared ledger, and every new participant strengthens the underlying network for all. Well-designed network tokens amplify this by aligning users, developers, and validators around growing the network, with value flowing back to contributors in a transparent feedback loop. This model is not new; companies from railroads and Standard Oil to Google, Meta, and AWS have thrived by inserting themselves into critical flows of value (goods, attention, compute). Financial markets make it even clearer: firms like Visa and major market makers generate immense revenue not by predicting markets but by being in the path of transactions. The combination of fund flow and network effects creates one of the most durable business structures. The high margins in traditional finance (payments, custody, lending, FX) represent prime targets. Crypto founders have the opportunity to build the next version—programmable, instant, global, and natively in the flow of funds. The frontier extends beyond finance to areas like computing/GPUs, AI training data, energy, robotics, and space—markets without entrenched intermediaries, ripe for building new, efficient value rails on programmable infrastructure. Founders should ask: Are you in the flow of funds today? Does your revenue scale 10x with the value of activity on your platform? Where in your target market are profit margins highest relative to value created? The opportunity is clear: embed your startup into the new flows of value and let the network effects accumulate.

marsbit06/12 02:38

a16z Crypto Partner: Cash Flow is the Moat

marsbit06/12 02:38

When AI Traffic Surpasses Humans, How Do You Prove You're Human?

As AI-generated web traffic now surpasses human activity, the internet's foundational business models—built on human attention, browsing, and advertising—face severe disruption. AI agents crawl websites at immense scale without generating ad revenue, while AI summaries divert traffic from original content sites. In response, over 2.5 million sites are blocking AI crawlers, and protections like Cloudflare's "honeypot" traps have emerged, though advanced AI can bypass these. The collapse of traditional CAPTCHAs, which assumed machines were weaker than humans, has led to a shift toward behavioral biometrics for human verification. Companies like IBM and BioCatch now analyze unique human patterns—cursor movements, typing rhythms, keystroke dynamics, and even cognitive delays like the Stroop effect—to distinguish real users from bots. These biometric signatures are difficult to fake or alter, offering a new layer of security but raising significant privacy concerns. Two competing visions for a reliable human verification system are emerging. One, exemplified by Sam Altman’s World (formerly Worldcoin), uses centralized iris scanning to generate unique credentials, though it faces bans and criticism over unauthorized data collection. The other employs cryptographic zero-knowledge proofs, allowing users to prove they are human without revealing identity or biometric data, as advocated by Vitalik Buterin. However, decentralized approaches risk exploitation through identity renting in economically unequal regions. The central dilemma is between a scalable but privacy-invasive centralized system that permanently controls users' biometric data, and a privacy-preserving cryptographic system vulnerable to real-world economic manipulation. The author expresses a preference for the cryptographic path, arguing that despite its flaws, it avoids the irreversible biometric surveillance inherent in centralized alternatives.

Foresight News06/11 09:14

When AI Traffic Surpasses Humans, How Do You Prove You're Human?

Foresight News06/11 09:14

Public Version of Mythos Officially Launched: Analyzing the Advantages and Limitations of AI Smart Contract Auditing

Publicly available Mythos, Anthropic's AI model, has officially launched, demonstrating both significant potential and limitations in smart contract security auditing. The article analyzes its capabilities through real-world cases. AI excels in identifying subtle, low-level vulnerabilities through pattern recognition and large-scale code screening. A key example is detecting a storage slot collision between a custom rewards mapping and a third-party library's ReentrancyGuard, a vulnerability easily missed in manual audits. In the recent Zcash incident, AI also rapidly discovered a critical soundness bug that had remained hidden for years. However, AI currently struggles with complex, interconnected scenarios. When tested on the Curve LlamaLend sDOLA exploit, which involved manipulating prices across multiple protocols (Curve pools, lending markets) to trigger liquidations, Fable 5 failed to identify the core cross-protocol attack vector. These scenarios require a deep understanding of DeFi economic models and multi-contract interactions. In conclusion, while AI tools like Mythos significantly boost efficiency in finding standardized, syntactic vulnerabilities, they cannot yet replace expert analysis for complex, business-logic, and cross-protocol attacks. An effective audit workflow combines AI's speed for initial screening with human expertise for in-depth, holistic analysis.

marsbit06/11 08:06

Public Version of Mythos Officially Launched: Analyzing the Advantages and Limitations of AI Smart Contract Auditing

marsbit06/11 08:06

Behind Musk and Huang Jen-hsun's 'AI Factories', an Unseen Battle for Freshwater Has Begun

Behind the "AI factories" of Elon Musk and Jensen Huang lies a hidden battle for a critical resource: fresh water. As AI models like ChatGPT and Claude process billions of prompts daily, they consume vast amounts of water for cooling. By 2030, global AI infrastructure is projected to use 9.3 trillion liters annually—enough to meet the basic needs of 1.3 billion people. This "water grab" stems from the massive heat generated by high-powered GPUs. Over 70% of data centers use evaporative cooling systems, where water absorbs heat and evaporates into the atmosphere, depleting local groundwater. Training models like GPT-4 can consume over 600 million liters of water. Tech giants like Google and Microsoft report skyrocketing water usage, sparking conflicts with local communities over resources. A flashpoint occurred in Memphis, Tennessee, where Musk's xAI built the Colossus supercomputer. It draws nearly 3.8 million liters of drinking water daily from local aquifers, leading to public outrage and legal action. In response, xAI is building an $80 million water recycling plant to use treated wastewater instead. Facing pressure, companies like Microsoft promote "waterless" closed-loop cooling systems. However, these systems increase electricity consumption by 20-30%, shifting the water burden to power plants, which require immense cooling water themselves—a case of indirect water footprint transfer. For China's AI industry, this crisis offers a strategic warning and opportunity. Instead of replicating the West's resource-intensive model, China can leverage its "East Data, West Computing" policy to locate data centers in cooler, water-rich regions like Guizhou. Furthermore, developing lightweight edge computing for smart homes and embodied AI robots can drastically reduce the need for constant cloud queries, cutting both water and energy consumption at the source. The freshwater war underscores a fundamental question: Will AI be a tool for human advancement or a silicon-based monster competing for our planet's last drops of clean water? The answer is becoming clearer as the water vapor rises.

marsbit06/11 05:23

Behind Musk and Huang Jen-hsun's 'AI Factories', an Unseen Battle for Freshwater Has Begun

marsbit06/11 05:23

AGI is Just One Step Away

The article discusses Anthropic's release of the Fable 5 model, a heavily restricted version of its powerful Mythos model. Initially unveiled in April, Mythos reportedly identified over 10,000 high-risk vulnerabilities for 50 enterprise clients, causing significant concern. Due to its dangerous capabilities in areas like autonomous cyber-attacks and biochemical weapons design guidance (classified as CB-1 level), the unaltered Mythos 5 remains limited to about 200 vetted entities like government agencies. Fable 5, released with a safety classifier, demonstrates extraordinary performance, leading benchmarks in coding (SWE-Bench Pro), software engineering, and research. It exhibits true "long-horizon agency," autonomously planning and executing complex, multi-step tasks like migrating 50 million lines of code in a day, moving beyond simple question-answering. The article positions Fable 5 at OpenAI's Level 3 ("Agent") and progressing toward Level 4 ("Innovator"), suggesting AGI (Artificial General Intelligence) is within reach, potentially 18-24 months away. To mitigate risks, Anthropic implemented a two-layer safety "cage": a silent routing system that redirects dangerous queries to a weaker model, and a mandatory 30-day data retention policy for all Mythos traffic to detect patterns of malicious use. Despite its high cost ($10/$50 per million input/output tokens), the model targets the enterprise market, where its unparalleled productivity and defensive capabilities against AI-powered cyber threats justify the premium. This signals a market maturation where top-tier AI becomes a strategic, high-value tool for businesses, potentially widening the gap with consumer-focused models and accelerating the rise of "one-person companies" while disrupting labor markets.

marsbit06/11 05:10

AGI is Just One Step Away

marsbit06/11 05:10

活动图片