ЦИФРОВОЙ ПОТОК [DRiver]
5 subscribers
31.8K photos
8 videos
173K links
Конвергенция блокчейна, искусственного интеллекта и квантовых технологий представляет собой передний край инноваций.
Download Telegram
chainyoda
RT @songadaymann: 6000 days.
6000 songs.

A story that almost broke me:

The $3 million I made and lost.
My $1.1 million tax bill.
The Autoglyph that saved me

This is SONG A DAY #6000

🌩️CRYPTO TAX NIGHTMARE🌩️

The journey continues below👇 https://t.co/ufUKNPiMIH
tweet
ЦИФРОВОЙ ПОТОК [DRiver]
Photo
chainyoda
RT @spectatorindex: UNITED STATES: Chances of Trump being impeached again are at 48% according to prediction market @Kalshi https://t.co/RVqctbki8F
tweet
chainyoda
RT @coopernicus01: BREAKING: Your cabal is not even the main cabal. It’s like 3 cabals down on the MLM scheme. You are getting 4th hand information in your cabal. You barely show up at the bottom of the pyramid.
tweet
ЦИФРОВОЙ ПОТОК [DRiver]
Photo
chainyoda
RT @Mira_Network: AI x Crypto News we're keeping up with this week:

@HuggingFace Unveils Two Open-Source Humanoid Robots

Following its acquisition of Pollen Robotics, the AI platform announces HopeJR, a full-size humanoid with 66 degrees of freedom capable of walking and arm movement, and Reachy Mini, a desktop unit for AI app testing. Both robots are fully open-source to prevent robotics from being dominated by black-box systems from major players.

@PluralisHQ Achieves Decentralized LLM Training Breakthrough

For the first time, researchers demonstrate LLMs can be split and trained across consumer devices over the internet without speed or performance loss. Their "Protocol Learning" approach places one transformer block per device and compresses inter-block communication by 100x while maintaining training dynamics. This discovery of low-stable rank properties in transformer output projections enables true community-owned models as a third path beyond closed and centralized open-weight releases.

@hyperbolic_labs Launches Claude GPU Rental Integration

The MCP server enables Claude to autonomously discover, rent, and manage GPU instances through natural language commands. Users can instruct Claude to rent H100s, SSH into instances, run commands like nvidia-smi, download models, execute training scripts, and terminate instances when complete.

@Google Releases AI Edge Gallery for Local Model Deployment

The experimental Android app enables users to download and run Hugging Face models locally without internet connectivity. Featuring capabilities like image analysis, AI chat, and code generation through models like Gemma 3n, the app includes a Prompt Lab with templates and configurable settings.

@Microsoft's Bing Offers Free Sora-Powered Video Generation

The first free access to OpenAI's Sora model arrives through Bing Video Creator, allowing users to generate 10 free 5-second videos in 9:16 vertical format. Additional videos cost 100 Microsoft Rewards points earned through Bing searches and Microsoft Store purchases. While generation can take hours even in "fast" mode, this marks a significant expansion of AI video tools.

@driaforall Introduces STIP Protocol for Privacy-Preserving Distributed Inference

The research introduces low-overhead, privacy-preserving distributed inference using the Secure Transformer Inference Protocol (STIP). Leveraging transformer architecture properties and permutation-equivariant operations, the protocol enables efficient privacy-preserving LLM inference at scale while supporting long-term security in decentralized networks.

The verification layer for trustless AI continues to evolve as innovations push the boundaries of what's possible in distributed computing, privacy preservation, and accessible AI deployment across both edge devices and decentralized networks.
tweet
ЦИФРОВОЙ ПОТОК [DRiver]
Photo
jesse.base.eth
RT @Archer_MD_: baseUSD @ 14% is the best way to put your USDC to work on @base. Base rate @ 9% and it can take size.

Enjoy your weekend and touch grass https://t.co/7kkhM8YN1h
tweet
chainyoda
RT @lex_node: unfortunately just having an entity that doesn't have equity holders and prohibits dividends does not magically create tokenholder 'alignment'--that's table stakes & every foundation (including the ones everybody hates) already does that

the real bare minimum should be:
-->governance docs are public
-->in those docs, the entity purposes are narrowly and carefully drafted to tailor to the particular community/token/protocol at issue (this is almost never the case), with few weasel words or "any other purpose permitted by law" clauses
-->there is some enforcement mechanism for actually aligned third parties (e.g., the DAO itself) to take legal action against the entity and its managers if they are violating the entity's rules

Very often, almost always, at least one of these elements is missing & the potential for loopholes and end-runs causing 'misalignment' with tokenholders is massive.

Just BORG it. 🖖😎
tweet