Note: The X account is not owned by the original Enron entity. Users should be aware of the risk.

Cathie Wood’s ARK Invest purchased approximately $9.1 million worth of Robinhood stocks on Wednesday, marking its largest single-day position increase since April 29 and its first purchase of the stock since September 8. The total purchase volume was 99,418 shares, allocated across three ETFs: ARKK, ARKW, and ARKF. According to data from BIT Stock Platform, Robinhood closed 3.3% lower on the day at $91.43, with its value down more than 23% over the past month. Earlier this month, ARK had sold a total of about $47.5 million in Robinhood shares between October 6 and 9. The purchase came after Robinhood released its financial report, which showed its third-quarter crypto trading revenue dropped 55% year-over-year to $120 million, while its prediction market revenue reached $173 million.

Digital asset and data center firm Galaxy Digital announced that its subsidiary Galaxy Digital LP has filed an application with the U.S. Office of the Comptroller of the Currency (OCC) to establish Galaxy National Bank, a national trust bank based in New York. If approved, the bank will operate under OCC oversight, providing digital asset custody, trading, staking, and fiduciary investment advisory services to clients nationwide. Galaxy founder and CEO Mike Novogratz noted that a national trust charter will help bring more business operations under a unified federal regulatory framework and expand customer services. The company added it currently holds over 50 licenses and regulatory approvals in the U.S., Europe, and Asia, including an SEC-registered broker, CFTC-registered swap dealer, and the recently approved New York State BitLicense and money transmitter license. The application remains pending OCC review.

Trade.xyz founder @sershokunin has published a lengthy article, announcing: "381 days after Trade.xyz’s first market went live on Hyperliquid, we are partnering with ICE. Upon CFTC approval, we will launch a dedicated U.S. market offering Hyperliquid-powered 24/7 perpetual contracts covering stocks, indices, commodities, forex, crypto, and other assets. U.S. users previously could not directly trade these markets; they will now gain access to innovative products paired with U.S. regulatory protection. ICE has made a strategic investment, and has invited participants including Stan Druckenmiller." In the post, @sershokunin also stressed the protocol’s vision: building a "universal exchange" that enables trading any asset anytime, anywhere. The ultimate goal is to realize the vision that "everything you own works in harmony with everything you trade" — such as using tokenized stocks as collateral, hedging indices, and trading pre-IPO companies, all completed within a single account.

DeFi protocol Curve Finance and the Stellar Development Foundation announced that Curve has launched on the Stellar mainnet, marking its first non-EVM native deployment. The StableSwap feature is now available, enabling users to permissionlessly create liquidity pools, swap assets, and provide liquidity. Curve founder Michael Egorov stated that the expansion to non-EVM chains is partially driven by institutional demand, adding that Stellar is relatively aligned with existing EVM usage habits. The project also plans to roll out FXSwap after an audit is completed, though no timeline has been provided. According to DeFiLlama data, Curve’s current total value locked stands at approximately $1.27 billion.

Intercontinental Exchange (ICE), parent company of the New York Stock Exchange, plans to launch a Hyperliquid-based perpetual futures market in the U.S. Following the news, HYPE saw a short-term rally, with large on-chain accounts building up positions en masse. According to TradingBeats' monitoring, within about half an hour after the announcement, 12 verified whales opened or rebuilt HYPE positions worth over $1 million each. Of these, 9 long positions totaled around $29.3173 million, while 3 short positions totaled around $8.0435 million, resulting in a long-to-short position value ratio of approximately 3.64:1, with a long net exposure of about $21.2738 million. The ratio of long to short accounts in this group is 3:1, with long positions accounting for roughly 78.47% of the total bilateral positions. The weighted average entry price for long positions is around $83.04, and for short positions around $82.86. Notably, an address starting with 0x9e2c opened a long position of 150,000 HYPE tokens, with an average entry price of approximately $82.47, a position value of about $12.7536 million, an unrealized profit of around $383,500, accounting for roughly 43.5% of the total long positions in this group.

Insight Beating AI Flash Briefing: Large language models can now learn language independently, but how text is split into tokens is usually determined by pre-trained tokenizers. This saves computing power, yet also constrains how models structure information. Fixed segmentation rules often fall short for unfamiliar words or code; skipping tokenization entirely to process raw bytes directly, however, leads to a sharp rise in computing costs. AI voice firm Cartesia introduced H-Net (Hierarchical Network) in July 2025 to resolve this tradeoff. The framework lets models read raw bytes directly and learn to segment and compress information autonomously. For example, an unfamiliar word may initially split into multiple segments; as it appears repeatedly in context, it gradually merges into fewer segments. This adjustment can occur during inference, with no retraining required. Cartesia has now unveiled a new H-Net version, with notable gains in training efficiency. In experiments with strictly controlled key conditions, computing power needed to hit the same validation loss dropped by ~22%, and computational efficiency for downstream evaluations nearly doubled. In larger-scale trials, the new H-Net matched the average score of open-source model SmolLM3 (trained on ~11 trillion tokens) across six tests using only ~1 trillion token-equivalent data, requiring roughly one-tenth the training compute of SmolLM3. This approach is not limited to text: the team trained H-Net on bacterial and archaeal genomes, finding that when processing human gene fragments, the model could automatically recognize the pattern of grouping every three bases in coding regions. Related DNA modeling experiments saw a 6–7x boost in computational efficiency. Unlike text, speech has no pre-defined word boundaries. Cartesia’s initial H-Net research was actually aimed at enabling voice models to learn phonemes and words directly from continuous audio, without relying on pre-designed segmentation rules.