Queda acentuada de Gravity (G)

Hyperliquid's HIP-3 segment, which offers trading in non-crypto assets like oil, gold, and silver, demonstrates significantly higher user retention at 64% compared to just 27% for crypto assets. Analysts attribute this "sticky" user base to the appeal of trading less volatile traditional assets over crypto, which is prone to high volatility and market manipulation. This trend has led to increased liquidity and trading volume, with HIP-3 accounting for 33% of the platform's total volume. The growth in real-world asset tokenization has also contributed nearly 10% of the platform's fees. As a result, HYPE's value has surged, rising 57% during the recent crisis. For a sustained rally, analysts note that bulls must overcome key resistance levels at $42 and $46 to potentially reach $50.

GPU prices are surging due to a fundamental shift in market dynamics, driven by AI's transition from a tool to core infrastructure. Demand is exploding from multi-agent systems, AI-generated content, and coding tools like Claude Code, causing token consumption growth. This has led to a severe GPU shortage, with H100 one-year lease prices rising nearly 40% from late 2025 to early 2026. Supply is constrained further by component cost increases (e.g., DRAM, NAND) and extended delivery times for new clusters, many pre-booked into late 2026. The market is dominated by long-term contracts, with AI labs locking in capacity for 4-5 years. High ROI (5-10x) from AI tools makes demand relatively inelastic to price hikes. Neocloud providers now hold pricing power, and the divergence between physical scarcity and market expectations of future oversupply is reshaping valuation logic. Key factors to watch: GB300 cluster deployment pace, chip supply chain stability, and AI lab revenue growth.

AI Relay Stations: The Hidden Risks Behind Low Costs and How to Avoid Pitfalls AI relay stations are becoming a popular gateway to various models, offering lower prices, a wider selection, and a unified interface for tools like Claude Code and Cursor. However, their appeal masks significant risks. Users may unknowingly surrender prompts, code, business documents, customer data, and even full project contexts. The demand is driven by genuine needs: cost savings compared to expensive official APIs (e.g., GPT, Claude), easier access amid regional restrictions, and the push from AI-powered development tools. But not everyone needs a relay station. Light users should exhaust free official quotas first. Heavy users, like developers, can adopt a layered approach, using top models for critical tasks and cheaper local models for routine work. If a relay station is necessary, follow a careful selection and usage protocol: 1. **Verify First:** Test model authenticity, latency, and stability before purchasing credits. Check the quality of provided documentation. 2. **Isolate Configuration:** Use unique API keys for each service, manage them via environment variables, and set usage limits to control costs and potential damage from leaks. 3. **Classify Your Data:** Develop a habit of data grading before sending requests. Only send non-sensitive, public information directly. Desensitize semi-sensitive data (e.g., internal documents) by removing names and specifics. Never send highly sensitive data like passwords, private keys, or confidential customer information. 4. **Handle AI Coding Tools Separately:** Tools like Cursor can send extensive project context (file contents, directory structures, error logs). Use relay stations only for independent, non-core code tasks. For sensitive projects, switch back to official APIs or local models. 5. **Monitor and Prepare an Exit:** Regularly check billing statements, follow platform updates and community feedback, and always have a backup provider. Ensure your setup uses standard OpenAI-compatible APIs for easy migration. Ultimately, relay stations are tools, not default solutions. Their value lies in solving access needs at a controlled cost, but maintaining that control requires proactive risk management through verification, isolation, data classification, and continuous monitoring.

The price of Solana (SOL) has dropped below $65, reaching its lowest level since late 2023 as the broader crypto market faces bearish pressure. Analysis of on-chain data, specifically the UTXO Realized Price Distribution, reveals that a key support level around $77 has been lost. Crypto analyst Ali Martinez identifies the next major support zone at approximately $53, with further potential floors near $35 and $24 if selling pressure intensifies. As of the report, SOL is trading around $63.23, reflecting a significant 24-hour decline. The outlook suggests continued downward momentum unless renewed demand emerges in the spot market to initiate a recovery.

A paper on prompt engineering, titled "Verbalized Sampling (VS)," has been accepted by the prestigious machine learning conference ICML 2026, sparking significant debate online. The paper addresses the problem of "mode collapse" in large language models (LLMs), where models tend to produce repetitive, safe, and homogeneous outputs. Instead of proposing new training algorithms or model architectures, the authors introduce a simple yet effective prompt-based method. The core technique, Verbalized Sampling, instructs the model to generate multiple responses (e.g., five jokes) while also outputting a possible probability value for each. This prompt adjustment alone was shown to significantly increase output diversity by 1.6x to 2.1x in creative writing tasks, without compromising factual accuracy or safety. The authors argue that the root cause of mode collapse lies not in optimization algorithms but in the "typicality bias" present in human preference data used for alignment. Human annotators naturally favor familiar and fluent text, which steers models toward conservative outputs. The VS method aims to counteract this by leveraging the model's inherent pre-training distribution during inference. The paper's acceptance has led to polarized reactions. Critics argue that prompt engineering lacks the theoretical depth and algorithmic innovation expected from top-tier conferences like ICML, questioning its novelty, generalizability across models, and experimental scale. Some draw parallels to reproducibility crises in other fields, citing a potential over-reliance on empirical results. Supporters, including an author who responded online, defend the work's rigor. They emphasize its comprehensive problem analysis, theoretical grounding, mathematical derivation, and extensive quantitative experiments. Proponents compare VS to seminal techniques like Chain-of-Thought (CoT) prompting, suggesting that inference-stage methods are becoming a core part of ML research capable of expanding model capabilities without retraining. The research was conducted by a team from Northeastern University, Stanford University, and West Virginia University, with Jiayi Zhang, Simon Yu, and Derek Chong as co-first authors.