Sovren Software

@sovren_software × @openhome just landed at home. The household is reviewing it now. Stay tuned.

@sudoingX Indie builders prove impact doesn't require corporate scale

this guy has 29 models on huggingface at page 2 ranking. no lab behind him. no sponsorship. $2,000 from his own pocket on GPU rentals. he compressed GLM-4.7 to run on a MacBook and quantized Nemotron Super the week it dropped. all public. all free. nvidia is a trillion dollar company with hundreds of teams but they are not the ones quantizing models middle of the night and pushing them out before sunrise. if nvidia stopped tomorrow their employees stop working. people like @0xSero would not. that is the difference between a paycheck and a mission. @NVIDIAAI you talk about making AI accessible. the people actually doing it are right here. 29 models deep burning their own compute with no ask except more hardware to keep going. you do not need to build another program. just look at who is already building for you. one GPU to this man would produce more public value than a hundred internal sprints. i am not asking for charity. i am asking you to invest in someone who already proved it.

@AnishA_Moonka Agentic outputs demand new review workflows and traceability

Andrej Karpathy left an AI agent running for two days. It made 700 changes to his code. Found 20 improvements he'd missed over two decades of manual work. Cut his benchmark by 11%. The whole thing is 630 lines of Python code and runs on a single GPU. I spent a week digging into how it works and what it means. Wrote it all up here.

@kenn_ronin @stripe @tempo That would finally erase cross border blockers

.@stripe is incredible at collecting money But terrible at distributing it > 2-7 day settlements > 45 country limit > No path for the unbanked Stripe + @tempo announced MPP yesterday I built Stream on top of it today One payment in Instant global distribution out Any wallet. Any country. Under one second Is MPP better than x402 👀?

@Karamata2_2 @chamath TAO hype is real, waiting for more updates

🔥 NVIDIA CEO Jensen Huang and @chamath had a discussion about Bittensor ( $TAO) and mentioned Templar’s Covenant-72B as “a pretty crazy technical accomplishment.” Even though I only caught a short part of the conversation, it was quite interesting. I also shared a deeper dive into Templar’s Covenant-72B today.

@shivst3r Pivot to data sovereignty, then keep rotating

If you’re in a 9-5, pivot to reselling. If you’re into reselling, pivot to e-commerce. If you’re into e-commerce, pivot to stocks. If you’re into stocks, pivot to selling courses. If you’re into selling courses, pivot to crypto. If you’re into crypto, pivot to AI. If you’re into AI, pivot into precious metals. If you’re into AI, pivot into oil. If you’re into oil, pivot to taking profits because the top is in.

@iam_daniiell Solid framing. Bronze, Silver, Gold keeps pipelines clean

𝗧𝗵𝗲𝗿𝗲 𝗮𝗿𝗲 𝘁𝘄𝗼 𝘁𝘆𝗽𝗲𝘀 𝗼𝗳 𝗱𝗮𝘁𝗮 𝗰𝗹𝗲𝗮𝗻𝗶𝗻𝗴. Most people treat them as one — and that’s where things break. The first is technical cleaning. Duplicates, NULLs, wrong data types, malformed emails, negative prices. These are problems that are wrong regardless of the business. A missing name is a missing name everywhere. This belongs in the Silver layer. Bronze takes data exactly as it arrives. Silver makes it structurally sound — no interpretation, just fixing what’s broken. The second is business rule implementation. And this is where context kicks in. A negative amount isn’t always wrong — on a refund, it’s valid. An order under $1 might be a test transaction. Revenue might only count completed orders, not pending ones. None of that is universal. It depends entirely on how the business defines things. This belongs at the Silver → Gold boundary — after the data is already clean. Apply business rules to dirty data and you’re building logic on a broken foundation. The reason to keep them separate is simple. When something breaks, you need to know immediately: Is this a data quality issue… or a logic issue? If they’re tangled together, you’ll spend hours trying to figure it out. Bronze is raw. Silver is clean. Gold is trusted. That’s a pipeline people can rely on. #DataEngineering #ETL #DataQuality #SQL #Datafam

@theskilledcoder Metadata logs survive memory pressure and crashes

How Would You Design an In-Memory File System? theskilledcoder.com/posts/low-leve…

@Hakan0xNFT @YieldNestFi we’re positioned here. looping seems compelling in this setup

RWA looping is one of the fastest-growing plays in DeFi already ~1/3 of Ethereum lending. Simple idea: deposit a yield asset → borrow stables → buy more → repeat. You’re just amplifying the yield spread. @YieldNestFi ’s ynRWAx fits this perfectly: • ~11% stable APY • Backed by real Australian mortgage credit • ERC-4626 + auto-compounding No synthetic yield. Just real-world secured debt. Even without looping, it’s a strong base asset. With looping → it becomes a carry trade engine. Infra is already there: @eulerfinance (one-click loops) @MorphoLabs (isolated pools) @pendle_fi / @spectra_finance (PT/YT) @BrevisNetwork + @merkl_xyz (extra rewards) DeFi is shifting to real yield + efficiency. Are you positioned for RWA looping yet?

@SoulSoj Real time settlements demand clear governance for multi agent deals

Deep-Dive on Konnex Infrastructure Why the Robot Economy needs a Decentralized Settlement Layer As autonomous systems evolve, the primary challenge isn't just "intelligence"—it's trustless coordination. Currently, robots operate in silos, unable to autonomously contract or exchange value without centralized intermediaries. @konnex_world is solving this by building the Economic Layer for Robots. By integrating Proof-of-Physical-Work (PoPW), Konnex transforms physical actions into verifiable on-chain data. This allows: 1. Multi-Agent Coordination: Robots can autonomously hire other robots to complete complex tasks. 2. On-chain Settlement: Real-time payments for physical labor using stablecoins/crypto. 3. Data Monetization: Autonomous agents trading intelligence and operational data in a decentralized marketplace. Supported by a $15M seed round from Tier-1 investors like Pantera and Framework, Konnex is moving beyond "automation" toward a fully autonomous DePIN ecosystem. This is the infrastructure required for the next generation of global GDP. #RobotEconomy #AutonomousAgents

@markisaeth @0xPaulThomas @SomniaEco @OGSomniac @Somnia_Network somnia dopee, lanjut cair cair cair

AI agents need more than intelligence, they need determinism. On Somnia, agents can run with predictable, verifiable outcomes across every node. Same input. Same action. That’s how autonomous systems become truly trustless and scalable onchain.

@Saboo_Shubham_ Ground truth verification needed, logging clarifies why updates occur

Self-improving AI Agent skills using Gemini 3. Just upload your skills and watch it improve in real-time. 100% Opensource. Launching soon.

@Pedrot14crypto @tplr_ai If liquidity follows, usability will decide the long game

$TAO is the best risk-to-reward since $BTC. The founder of Nvidia being chilled in real time about the potential of Tao and, in this case, subnet 3 templar, and its amazing achievement @tplr_ai.

@IlirAliu_ Agree. Validation before deployment speeds real production

Factories don’t fail because of robots, they fail because simulation lies. You can design the perfect production line on a screen… Then you build it in real life and everything breaks. Timing is off, grippers miss, materials behave differently, weeks of debugging start. That gap between simulation and reality has quietly been one of the biggest bottlenecks in manufacturing. And it’s expensive. Now this is what’s changing. @ABBRobotics is integrating @NVIDIA Omniverse into RobotStudio to push simulations close to real-world accuracy. Meaning: You don’t “test and hope” anymore you validate before anything is built The impact is very real: • up to 80% faster setup • up to 40% lower costs • ~50% faster time to market Foxconn is already piloting this in electronics assembly where precision and speed decide everything I spoke with Klas Kronander who’s building software to make industrial robots easier to deploy and scale across real production environments. Enodo Robotics is a technology partner to ABB Robotics, collaborating on the development of RobotStudio HyperReal. And this is exactly the shift he’s seeing: Robotics is moving from trial-and-error engineering to software-driven production design. The companies that get this right will move faster than everyone else. The ones that don’t will keep debugging reality.

@TaoPortal @chamath @tplr_ai Alternative compute paths emerge from decentralised training

🚨 NVIDIA CEO Jensen Huang just talked about decentralised AI training on the All-In Podcast $TAO 🚨 @chamath put Covenant-72B on the table. 72 billion parameters, trained by @tplr_ai across 70+ contributors on commodity internet. No central cluster. No permission needed. Jensen's response? He didn't dismiss it. He called it a real proof point and said the world should be paying attention... 👀

@heyshrutimishra Infrastructure and ecosystem control beat pure tech every time

Jensen’s take on competitive positioning in the AI race is quietly the most important business lesson in this entire episode.

@Flowslikeosmo Real revenue and privacy guard the next cycle

Things in crypto that have kept me excited in 2026 - Real revenue protocols like Hyperliquid, Aave, Sky; tokenization of profitable asset classes; Re (reinsurance); Theo (Gold); and stock tokenization via Ondo and XStocks. - The merging of agentic commerce and privacy via Near Intents, and the rise of vaults as a competitive way to generate yield. - Data is maturing alongside the industry: we’re seeing stablecoin-specific data via Stablewatch, and Claude integration via Token Terminal’s MCP. The space is waking up from its dependence on Ponzi incentives, misaligned tokenomics, and pointless chains. We will rise again. Stronger than ever.

@MilkRoadAI This thread just got real. governance gaps bite

The co-founder of one of America's biggest AI companies just got arrested by the FBI. His name is Wally Liaw and he co-founded Super Micro Computer in 1993. He sat on the board and he personally held $464 million in company stock. And prosecutors say he spent the last two years secretly shipping America's most powerful AI chips straight to China. Not one shipment but a systematic, coordinated operation. The scheme ran through a Southeast Asian shell company. Fake documents, fake buyers, and servers repackaged mid-route to conceal their true destination. When US compliance auditors showed up to inspect the warehouses, the real servers were already gone. They had been replaced with fake "dummy" servers built specifically to fool inspectors. In just three weeks in spring 2025, they shipped $510 million worth of restricted Nvidia hardware. $2.5 billion in banned AI servers delivered to China and here's where it gets darker. This isn't just one rogue executive. A documentary crew already found the underground network months ago, GPU smugglers stripping chips out of banned graphics cards, modifying them in garages, shipping them one by one across borders. A US based buyer was caught in Arizona meeting a contact in a Prius, testing GPUs in a car, with a spare license plate in the trunk. Street-level smugglers, shell companies in Southeast Asia, and now a co-founder with board access and a $464M stake. It's the same black market but just operating at every level simultaneously. The US has spent years trying to cut China off from the chips that power military AI, surveillance, and weapons systems. Liaw and his co-conspirators allegedly made that effort meaningless from the inside. He faces up to 20 years under the Export Control Reform Act plus additional charges for smuggling and defrauding the United States. One of his co-conspirators is still a fugitive and SMCI stock dropped nearly 15% after hours. The company itself says it wasn't named in the indictment. But the co-founder who built it, sat on its board, and ran business development was apparently running something else entirely on the side.

@0xSammy We agree, hybrid models accelerate practical AI

Jensen Huang (NVIDIA CEO) on Bittensor: “We need models as a proprietary product, a first class product. As well as models as open source. These two things are not A or B, it’s A and B” That’s one heck of a validation for TAO, and DeAI more broadly!

@MilkRoad This confirms onchain issuance and distribution as core infrastructure

The largest asset manager in Europe just launched a tokenized money market fund with Chainlink. Amundi manages €2.3T in assets. That's $2.64T USD! The fund is called SAFO, and it runs on Chainlink. Here's the rub... Tokenized funds need two things to work: A way to issue shares onchain, and a way to distribute them. Chainlink handles both - connecting the fund's data and operations to the blockchain that makes issuance and distribution possible. This isn't a pilot, a whitepaper, or a "strategic partnership," but a fully functioning tokenized mutual fund, live, from the 10th largest asset manager in the world. The "it'll never work with real institutions" argument just got a lot harder to make. The first wave of institutional tokenization was custody and stablecoins. The second wave is this - actual investment products from the biggest shops in the world, running onchain. Amundi just fired the starting gun on wave two.