🚨 Anthropic committed up to 1M TPU chips for... Claude. Openai is leasing TPUs for chatgpt inference. Here's How kernels work on TPUs (deep dive 2/6 by emi) pallas is Google's answer to kernel writing. a python kernel SDK built on JAX. still very experimental (jax.experimental.pallas). on TPU it compiles through mosaic; on GPU it lowers to triton. if you know CUDA, the syntax will feel familiar but the execution model is completely different. in CUDA, grid=(4,4) launches 16 blocks running simultaneously across SMs. in pallas, those 16 iterations run one after another in lexicographic order. no threads. no warps. no blocks. no occupancy tuning. a TPU is a sequential machine with a very wide vector register — more like a CPU than a GPU. performance comes from width: a 128x128 systolic array doing matmul and an 8x128 SIMD vector unit doing everything else. maximum parallelism on chip: 2, one per TensorCore in megacore mode. three concepts replace CUDA's thread/block/grid hierarchy. Refs are mutable memory references. because execution is sequential, each iteration safely accumulates without atomics. in CUDA you'd need atomics or a separate reduction pass. the memory model is also very different from NVIDIA's. zero hardware caches. VMEM is 32-128 MiB of software-managed scratchpad — 500-1000x larger than GPU shared memory per SM. all data must be explicitly DMA'd from HBM to VMEM before any computation touches it. four levels: HBM → VMEM → VREGs → MXU/VPU, plus SMEM for scalar control data. every byte of data movement is your responsibility. this is like CUDA shared memory except it's 500x bigger and there's no cache fallback. pipelining is mandatory. without double-buffering HBM→VMEM transfers, the MXU just stalls waiting for data. this is the single most important optimization on TPU. and because grid execution is sequential and deterministic, consecutive iterations that need the same input block skip the redundant HBM transfer automatically, impossible on GPU where block execution order is undefined. the compilation pipeline is unlike anything in this series: python → jaxpr → stableHLO → XLA HLO (71+ optimization passes) → LLO (78+ passes) → 322-bit VLIW bundles. the compiler packs instructions for scalar, vector, matrix, and DMA units into a single 322-bit word. everything in that bundle executes in parallel, with no runtime scheduling.show more

First fully ML-framework-free 3D Gaussian Splatting implementation in LichtFeld... Studio. I’ve completed the migration of the full training pipeline to a custom CUDA-based tensor library. No PyTorch, no LibTorch, no autograd. Every gradient is implemented by hand, either through CUDA kernels or minimal abstractions on top. This makes it the first full training setup for 3D Gaussian Splatting with zero dependencies on existing ML frameworks. It’s not just about independence, it's about control! We now manage every byte of GPU memory, which opens the door to tighter optimization and finer performance tuning. The framework footprint is minimal, without pulling in gigabytes of ML runtime code that was never designed for real-time or graphics-driven applications. A few modules, such as the metrics and 3DGUT interfaces, are still being ported, and some operations are temporarily naïve, so performance is not yet on par with master. But this refactor lays the groundwork for: - A fully self-contained binary - Fine-grained memory optimization - Easier experimentation without the weight of an ML stack We’re getting close.show more

Day 11/90 of Inference Engineering How does vLLM work... and how is it used in production? Before we discuss how vLLM works internally, it helps to understand what vLLM is. At a high level, vLLM is an inference engine that is designed to serve LLMs to thousands of concurrent users efficiently while managing scarce compute and memory. The goal for vLLM is to maximize throughput and minimize latency; optimizing for the best inference economics and experience for end users. With every request from the end user, it eventually ends up in the engine core, gets scheduled alongside other requests from other concurrent users, executes on the GPU, and updates the KV cache with the new key and value vectors, and streams the tokens back to the user. The Scheduler decides what requests should execute next while continuously batching requests together to maximize GPU utilization. Continuous batching is an inference optimization that allows new requests to join a running batch as other requests finish generating tokens. This helps with keeping the GPU utilization high instead of letting it sit idle waiting for an entire batch to complete generating. After the scheduler dispatches the selected batch to the Model Executor, the Model Executor prepares the tensors and metadata required for inference, retrieves each request’s block table from KV Cache Manager, launches the optimized transformer forward pass on the GPU, computes the logits, updates the KV cache with the new key and value vectors, and finally returns the results for sampling and streaming. The KV Cache Manager uses the PagedAttention memory layout to allocate fixed-size cache blocks on demand and maintains a Free Block Queue on the CPU that tracks which blocks in the GPU’s Paged KV Cache are currently free. When a request needs additional KV cache space, the KV Cache manager takes a free block from the queue and assigns it to that request, thus avoiding an expensive search through GPU memory for available cache blocks. All of these components form the core of vLLM’s inference engine. The Scheduler determines what requests are executed, the Model Executor determines how those requests are executed, the KV Cache Manager determines where each request’s KV cache lives using the PagedAttention Memory Layout. This architecture enables vLLM to serve thousands of concurrent requests with high throughput, low latency, and efficient GPU memory utilization. Heres a little animation that visualizes everything! - I've also completed the forward pass for my mnist.c project. I had a nice chat with shrey birmiwal, such a knowledgeable guy. Excited to learn more about vLLM and implement a tiny-vLLM one day.show more

Researchers made KMeans 200x faster. And the new technique... also beats approaches like cuML and FAISS. Flash-KMeans is an IO-aware implementation of exact KMeans that redesigns the algorithm around modern GPU bottlenecks. By attacking the memory bottlenecks directly, Flash-KMeans achieves: - 33x speedup over cuML - 200x speedup over FAISS This speedup comes from how it moves through GPU memory. Standard KMeans runs in two steps, and both are bottlenecked by reads and writes to GPU memory: 1) The first step matches every point to its nearest centroid. Standard KMeans computes the full point-to-centroid distance matrix, writes it out to GPU memory, then reads it back to find each nearest centroid. That write-then-read round trip is the bottleneck. Flash-KMeans combines the distance calculation with the nearest-centroid step, so the result is computed on-chip and the full matrix is never written out. 2) The second step recomputes each centroid by averaging the points assigned to it. Standard KMeans has thousands of threads writing into the same centroid slots at once, so they stall waiting for their turn. Flash-KMeans sorts points by cluster first, turning scattered writes into sequential reductions that read and write memory in one efficient pass. Using these two optimizations at the million-scale, Flash-KMeans completes a standard KMeans iteration in a few milliseconds. The video below depicts this in action. Several reasons why this is important: KMeans has always been an offline primitive. Something you run once to preprocess data and move on. These speedups make the approach viable in several runtime-critical systems. ↳ Vector indices like FAISS use KMeans to build search indices. Faster KMeans means you can re-index dynamically as data changes. ↳ LLM quantization methods need KMeans to find optimal weight codebooks, per layer, repeatedly. What takes hours could now take minutes. ↳ MoE models need fast token routing at inference time. Flash-KMeans makes it viable to run this inside the inference loop, not just in preprocessing. I have shared the paper in the replies. That said, memory is the real constraint Flash-KMeans solves, and the problem is not just limited to clustering. The vectors a RAG system stores after indexing create similar bottlenecks. I wrote a detailed walkthrough recently on cutting this vector memory by 32x with binary quantization, querying 36M+ vectors in a few milliseconds. Read it below.show more

In 2025, the AgentFlayer exploit highlighted a new category... of risk in AI systems. It was not a traditional breach involving stolen credentials or broken encryption. Instead, it demonstrated how an autonomous AI agent could be manipulated into executing unintended actions by processing malicious instructions embedded inside content it automatically processes. The incident did not expose a flaw in one specific integration. It revealed a structural weakness in how many modern AI agents are built. Today’s agents are no longer passive language models. They read documents automatically, scan emails, connect to SaaS tools, access cloud storage, and execute actions across multiple systems. To be useful, they are granted meaningful permissions. That capability creates value, but it also expands the attack surface. Most agent environments operate in a trusted, plaintext execution model. Data is encrypted at rest and in transit, but it is typically decrypted during inference so the model can process it. That runtime visibility is where potential risk lies. In a zero-click scenario like AgentFlayer, an attacker can embed hidden instructions inside a document that the AI processes automatically. Because the agent may have access to connected systems such as Google Drive, Slack, or GitHub, it can potentially be influenced to retrieve sensitive information or perform unintended actions. The user does not need to click a malicious link or approve a suspicious request. Therefore, the core issue is that during execution, the system may have access to sensitive data and broad privileges, meaning whoever controls the execution environment ultimately controls access to that data. Now consider a different architectural approach. If a system is designed so that data remains protected during execution, the risk profile changes. On Nesa, privacy is enforced at the execution layer through Equivariant Encryption. Computation can occur on encrypted data, reducing the visibility surface during runtime. Sensitive inputs and models do not need to be exposed in plain text to infrastructure operators for inference to occur. This does not eliminate prompt injection, logic manipulation, or tool misuse. Encryption alone cannot prevent an agent from being instructed to take an unintended action if it has been granted that permission. What it does do is materially reduce confidentiality risk. By limiting access to readable sensitive data during execution and reducing unilateral visibility at the infrastructure layer, the potential blast radius of a successful manipulation attempt is constrained. As AI agents become more autonomous and embedded into enterprise workflows, security must move deeper into architecture. The goal is not to claim invulnerability. It is to reduce trust concentration and contain systemic exposure when failures occur. AgentFlayer was not simply a one-off exploit. It was a reminder that in autonomous systems, execution-layer design determines how risk propagates.show more

OpenClaw, but built for normal people. Sim is an... open-source platform that lets you build AI agent workflows on a drag-and-drop canvas. Connect them to channels like Telegram and WhatsApp and deploy without writing a single line of code. They also have a built-in Copilot that generates entire workflows from plain English, which you can then tweak and customize in the UI. Key features: - Free and open-source (Apache 2.0) - Vector store integration for RAG-grounded agents - Self-host with one command (`npx simstudio`) - Run fully local with Ollama, no API keys needed - Supports vLLM for production-grade self-hosted inference The thing I really like about Sim is the level of control you get. You can add conditional branching, parallel execution, human-in-the-loop approval gates, and even nest workflows inside other workflows. Everything is visible on the canvas, so you know exactly what your agent is doing at every step. And you can build a workflow in Sim, deploy it as an MCP server, and plug it into any agent, including OpenClaw. I've shared the link to Sim's GitHub repo in the next tweet.show more

The largest theft in history has already happened. The... people behind it just cannot open what they stole yet. Right now, intelligence agencies and criminal groups are quietly copying the world's encrypted data, bank records, medical files, state secrets, private messages, and storing every byte untouched. They cannot read any of it. They are collecting it anyway, because they know the key is about to be invented. The strategy has a name, harvest now, decrypt later, and in 2026 it stopped being theory. Washington declared this the Year of Quantum Security in January, backed by the FBI, the NSA, and NIST. Canada ordered every federal agency to file a migration plan by April. Europe set its deadline for December. Governments do not impose operational deadlines on a someday problem. They do it when the clock is already running. Here is what moved the clock. Every password, every transfer, every secret on Earth is protected by one assumption, that a certain math problem is too hard to solve. Quantum computers solve exactly that problem. For years the machine that could do it looked decades away. Then in late 2025 Google's Willow chip cracked the hardest part of building one, and in March 2026 Google's own researchers estimated that breaking the encryption behind Bitcoin might take fewer than 500,000 qubits, down from 20 million, and could run in minutes. The day this becomes real has a name, Q-Day, and the latest estimates place it between 2030 and 2033. Now make it concrete. Roughly 6.5 million Bitcoin, about a third of every coin that will ever exist, worth close to 500 billion dollars, sit in addresses that have already exposed the very key a quantum computer needs. That includes the coins of Satoshi, the anonymous creator. On Q-Day they become, in the researchers' own word, trivially stealable. It would not look like a crash or a whale selling. It would look like half a trillion dollars of the most secure money ever built simply walking out the door. The asset designed to trust no one and no institution turns out to rest on a single unverified bet, that one math problem stays hard forever. This is what sits beneath the entire digital world. A bank balance, a Bitcoin, a classified cable, all of it is real only because of a proof you supposedly cannot forge. Quantum breaks the proof. Everything we call secure is true only until someone finally checks, and for the first time the check is visible on the horizon. You cannot know whether your data has already been copied. You cannot know the exact day the key arrives. The trust holding up the digital age is a clock counting down to a zero no one can see. The honest counter matters. No machine on Earth can break this encryption today, and serious cryptographers still argue the real threat is a decade or more away. The timeline is far from certain. Quantum-safe codes already exist, the migration has started, and Bitcoin can move its coins to safety before Q-Day if it acts in time. The danger is not that everything breaks tomorrow. It is that anything which must stay secret into the 2030s, a state secret, an identity, a private key, is being stolen today and is already on the clock. The breach is not coming. It is already here, sitting in storage, perfectly encrypted, waiting for a machine that does not exist yet to read it out loud. Research and opinion, not investment advice.show more

Free NVIDIA GPU with 16 GB VRAM GPU for... Running Local LLMs! If you want to master local LLMs but you're waiting until you can afford a $1,500 GPU, you're honestly not going to make it. The open source AI ecosystem is moving way too fast for you to wait on your budget to catch up. Especially when you can build a bleeding edge inference engine from scratch right now, completely for free. You don't need a heavy local rig to start. Google is literally letting you use an enterprise grade NVIDIA Tesla T4 GPU for $0/hour. At standard cloud computing rates (~$0.20/hr), Google Colab’s 4 hour daily free tier hands you roughly $24 worth of data center tier GPU compute every single month. And most people just waste it. Let’s talk about the hardware you get access to for free. The NVIDIA Tesla T4 is an absolute workhorse: - Architecture: NVIDIA Turing (TU104) - VRAM: 16GB GDDR6 (320 GB/s bandwidth) - Compute: 320 Tensor Cores | 2560 CUDA Cores - Performance: 130 TOPS INT8 | 8.1 TFLOPS FP32 - Power: Sipping energy at a max 70W TDP This is the exact same hardware I used to run DeepMind's Gemma 4 26B A4B QAT MoE at a 250,000 context window without a single Out Of Memory (OOM) crash. If you have a web browser and 10 minutes, you have everything you need. I’ve put together a fully documented, cell by cell Google Colab notebook that teaches you exactly how to do this. Here is what the notebook actually teaches you: - How to provision an Ubuntu Linux environment with CUDA 13.0 and verify your driver stack. - How to pull the source code and compile the latest llama.cpp C++ binaries from scratch, specifically optimizing the build for your exact GPU using the -DCMAKE_CUDA_ARCHITECTURES=native flag. - How to directly download quantized local LLMs (GGUF format) straight from HuggingFace using the CLI. - How to manage 16GB VRAM limits, offload neural network layers to the GPU, and push massive context windows. Compile raw llama.cpp, ollama run a model, or spin up the LM Studio CLI. Pick whatever stack you are comfortable with. just start building. No hardware. No credit card. No excuses. Bookmark this post right now so you don't lose the tutorial. Even if you don't have time to run it today, you are going to want this workflow in your engineering toolkit. The link to the free Colab Notebook is in the comments below. Lemme know if you need more tutorials like this.show more

HTML Artifacts are a big part of how I... work with agents now. Artifacts can be more than just static files. When combined with agents, they can take action or help you take action. This unlocks all kinds of interesting ways to work with agents. This is clearly the future. Check out this writing and scheduler artifact I built in a few minutes. It uses a bit of HTML and JS. All the data is in markdown (Obsidian vaults), so the agent can access and modify it at any time. No DB needed. No sophisticated functionalities. The agent decides all that for me based on the skills, context, and memory it has access to. The best part about this simple stack is that all the important information stays with me. This has allowed me to build a recursive self-improving system and automations that can better tap into coding agents like Codex or Claude Code. I could have paid or built an entire app for scheduling posts, and there are so many of them out there. But I don't need to. I've realized a simple artifact does the job. And the simplicity of it is actually an advantage. Very little maintenance for very high returns on personalization, time, and efficiency. The other benefit of this is that I can add features as I please. That level of personalization feels magical, and we should all be pursuing more of it. All of this just keeps compounding. Of course, this example is just about writing. But I have similar artifacts for research, design, experimentation, evaluation, and so much more. And no, I didn't actually publish the post example I shared in the clip. It was just for demonstration purposes. I actually spend more time than this when writing together with agents. Lastly, having built my own agent orchestrator tool has made me realize that simplifying the tool stack is a superpower. If you are curious about how all this works, I will do a live session next week:show more

I am a Licensed Electrical Contractor, I work for... a company that manufactures and installs commercial LED fixtures. Just finished a job for a US Govco building. 1. The drivers work on any voltage range between 115-277 or 460-480, depending on the application. 2. Flickering (Strobing) is 95% from a Defective Driver. 3. There is no possible way all these drivers are going bad at the same time. 4. The only logical, technical answer to this is that the frequency of the power is unstable. US is 60Hz European is 50Hz. 60 or 50 Cycles per Second. For this to happen on a large scale, the Utility Provider has to have an imbalance between Production and Load. Very uncommon. 5. Excessive loads on the grid, normally the computer switches part of the load to a different substation. 6. There isn't a logical explanation for this on a wide scale, other than more planned PSYOP'S, to create more panic and fear. 7. Walmarts, Airports, Hospitals, Essential Structures like Police, Fire, etc. Are always fed from a minimum of 2 different substations. Many are fed from 3 with automatic transfer gear. 8. If this wasn't a PSYOP, or Intentional planned exercise to create more panic. Utility workers would be all over this, in less than 30 minutes. If it's doing that to the lighting, it's happening to motors, compressors, and computers. Watch the left hand so you don't pay attention to the right hand.show more

$IREN "we haven't disclosed the specific amount of GPUs"... 1. 🤮 reminds me of $NBIS 2. Setting a terrible precedent here for future deals 3. Making it purposely difficult, to not let analysts properly value your 2027 revenue 4. Increasing the polarized view on IREN by the market However: "approximately 60MW of air-cooled Blackwells" 1. You typically don't talk about gross capacity in a deployment like this 2. If it would be gross capacity, the GPU hour rate at IT level would be crazy high (at PUE 1.2, $680m / 50 = 13.6m/MW) 3. At 60MW IT load, and ~14kW draw at DGX server level, we can get to ~4,286 DGX systems with 8 GPUs per. 4. Based on this we can conclude that 60MW of IT load can run approximately 34k DGX B300. 5. 34k DGX B300 at $680m/yr, would represent a GPU hour price of $2.28 Now this is the problem with not disclosing your GPU quantity. You purposely make your business model look bad, because by approach, you get to a GPU hour price that would imply a payback period of 4 years, where only the last year of the contract is 100% margin. But of course, we can also take "the glass is half full" approach. IREN has ordered 50K B300s from Dell. They have 2 purchase orders for this, 1 between Dell Canada and IE CA Leasing Ltd for 4 phases, and 1 between Dell USA and IE US Hardware 1 Inc (amended from IE US Hardware 4 Inc on April 27, 2026). The order for Canada is divided in 4 phases, and are going to Mackenzie for 80MW of gross capacity, which happens to be 4 buildings of 20MW. The order for Childress is divided in 2 phases, and are going to DC35 and DC36, (as depicted in the earnings presentation) and those are 50MW gross. The purchase price of the order for Childress was $1.2B, and for Canada it was $2.3B If we go with 50,000 B300s for a total of $3.5B then $1.2 would represent 34.285% of the 50,000 GPUs, or 17,140 B300s rounded down. For this calculation I will consider that $IREN will deploy 17,140 GPUs in 50MW gross capacity in DC35 and DC36 of block 3 in Childress.. That would imply at 1.2 PUE, IREN can run 17,140 B300s in 41.67MW IT load. Now by that ratio, they can run 24,680 GPUs in 60MW IT load — a massive difference with 34k units through the Nvidia DGX reference calculation. If common sense is applied, you can still get to 2 completely different outcomes, that show a difference of more than 9k GPUs. The GPU hour rate at 24.68k GPUs would be $3.145 per B300, as MASSIVE difference from the earlier calculated $2.28. Sure, the DGX system may be a factor here. And I'm sure that the reality is somewhere in the middle. But I personally hate this as an investor, to be unable to calculate profitability on unit economic basis. After all, contracts are signed on a $/GPU hour basis. Why hide this from your investors? Not being able to calculate payback periods, unable to calculate ROIC. And most importantly, we cannot properly assess the $NVDA deal on a contract basis. I really hope the payback period of this contract is not 4 years. I want the glass to be half full, but by starting to censor the purchases, IREN is taking a step in the wrong direction. Not a fan of this.show more

"PRICE IS WHAT YOU PAY. VALUE IS WHAT YOU... GET." I keep buying $Kekec and I have a strong conviction. Here's Why: While the market is down, and Kekec is declining with it, there are data points that few are considering. Kekec borned in October and since then has been posting a different and original 30-second video every day, which I find extremely funny. For the past couple of months, they have also been posting daily on Instagram, and the attention on Kekec (which doesn't present itself on social media as a memecoin) is growing, moreover, it's increasing exponentially. The number of followers is increasing by about 500-1000 a day. This is largely due to the fact that they are not just focused on the main account but have several others that post reels and redirect to the main one. In short, an excellent strategy to keep growing more and more. Instagram link: Guess What? Not only are the followers increasing, but the team's workload is also growing. In fact, for a little over a month, they have also started pushing on YouTube, and the data here is promising as well. YouTube link: If we want to make a comparison, we can take Pudgy Penguins as an example, which has shown it can reach millions and millions of users without mentioning that they are a WEB3 company that owns an NFT collection. Or, if we want to be more appropriate by comparing one memecoin to another, we could take PONKE. Thanks to the use of social media and the quality of their content, they managed to achieve incredible numbers, which then translated into an increase in the coin's price. Kekec came before PONKE, but that doesn't necessarily mean it's better than PONKE. I believe PONKE is unbeatable in terms of content, but I want to make you reflect on an important point. PONKE came after KEKEC, and after PONKE's success, many coins have emerged trying to imitate it. One of KEKEC's strengths, in my opinion, is precisely the fact that it leverages social media without being a copy-paste. Instead, it is a unique meme derived from a 90's film, and it uses a unique form of content. In short, KEKEC > KEKEC and no one else. I want to conclude by suggesting you follow them on Instagram and evaluate not only the exponential growth of their followers day by day but also observe how the views of each reel increase accordingly. Pay special attention to the comments. Many of the people commenting have no idea what it is, and you can see from the comments how Kekec generates particular emotions in people—strange but still emotions. Personally, I believe that when something is unique and even very strange, it needs time to be adopted. However, once it happens, it usually explodes and spreads like never before. A few days ago, a Kekec video was posted by a very popular meme page. They probably don't know what Kekec is about but thought the video could spark interest among their followers. How many other pages will do the same? Lastly, but not least, I want to point out how Kekec maintains a good market cap despite everything that has happened in the crypto world since October 2023. As far as I know and have personally observed, everything is extremely organic. There is no cabal behind it, and the quality is not reflected in a single jpeg but in work that has been ongoing daily for months. Every day they work harder, and the quality of their videos grows as well. I have no affiliations with the team, but I believe that Kekec truly deserves more in this world where we push celebrity or cabal-backed coins to hundreds of millions in market cap. I keep buying because the numbers suggest so. Don't just evaluate the chart (price), evaluate the data (value). BÂLKÂN DWÂRFshow more

Elon Musk gave the entire entertainment industry its expiration... date, and he is the one building the thing that kills it. Musk: “My guess is that we see the first compelling half hour, pure AI show next year.” Next year. A complete show generated entirely by AI. No writers. No actors. No cameras. No sets. No crew. No studio. Just a prompt and enough compute to render a reality that never physically existed. And shows are the easy part. Musk: “I say probably we’re maybe three years away from AI does the whole video game.” A show plays the same way every time. A game has to generate a living world that reacts to every decision in real time across every single frame. That is a fundamentally harder class of problem. And Musk put three years on it. Right now a single AAA title takes seven years and half a billion dollars across thousands of engineers and artists just to ship it. Musk is describing a world where one person types a paragraph and gets something comparable. The entire value proposition of a multi-billion dollar industry lives inside that gap. And it closes in thirty-six months. But the prediction is not the story. The person making it is. This is not an analyst speculating from the sidelines. This is the man building the largest AI compute clusters on the planet. The man who built xAI from zero in under two years. The man stacking hundreds of thousands of GPUs into facilities designed to do exactly what he is describing. When Musk says three years, he is not guessing about what someone else might eventually ship. He is reading you a delivery date off his own roadmap. Every media company on Earth is valued on a single assumption. That quality content is expensive and difficult to produce at scale. That one assumption is the structural foundation underneath every studio, every network, and every publisher in existence. Musk is dismantling it with raw compute. The studios still parading thousand-person production teams are not demonstrating strength. They are advertising the exact cost structure that one person with a prompt and a GPU allocation is about to make irrelevant. And it does not stop at entertainment. If AI can generate an interactive world that responds to human input in real time, it can generate anything. Advertising. Architecture. Training simulations. Product design. Every industry built on humans manually constructing visual experiences frame by frame is sitting on the same countdown Musk just read out loud. Now zoom out. Because this is not just an industry story. For the entire history of human civilization, the distance between imagining a world and actually creating one required thousands of people, millions of hours, and billions of dollars. That distance built Hollywood. That distance built the gaming industry. That distance made content scarce and studios powerful. Musk is collapsing that distance to zero. When the gap between imagining something and it existing disappears, every business model built on the difficulty of creation disappears with it. That is not disruption. That is a full inversion of how human beings create. Musk did not make a casual prediction on that podcast. He told you what he is building. He told you the timeline. And he told you which industries do not survive it. The entertainment industry is still debating whether this future is real. Musk is not part of that debate. He is building. And he just told you the delivery date.show more

Stateless History Node is almost like a regular Ethereum... node, but it doesn't store state and it doesn't have EVM execution. It's used only for syncing events and thus - is faster and gives you FREE INDEXING. You don't have to pay 6 figures for RPC anymore! Just spin up a Stateless History Node, plug rindexer or Ponder there, and enjoy free (AND FAST!!) indexing! This node is syncing >1000 blocks per second at my local pc (less than 6hrs for the whole Ethereum), and it should use less than 200GB - which means you can host it on a MacMini, Hetzner or whatever. You can futhermore filter that by using block ranges or bloom filters, etc - I haven't developed this yet. What you see is a proof of concept. It works via native devp2p 'eth' protocol, but with EIP4444 and The Prune we would have to also support era1 archives and Portal Network. But so far it works - there are plenty of peers serving historical receipts, and they serve them FAST! If you run Stateless History Node you can also serve the blocks and receipts - so that could help to preserve archival data too. For now there is no data validation yet (and even no data storage - that's a very early PoC), but we can verify validity of chain by simultaneously running a lightweight CL node (or not lightweight if you're extremely paranoid). And then support verifying the hashes of receipts and blocks with their parents, maintaining full integrity and zero trust. It's also written in rust, btw. So, I guess, at least for Ethereum Mainnet the era of RPC's pumping moneybags is over - there's finally a local, trustless and free indexing alternative available. Too sad this won't work for Optimism / Base , cause despite introducing P2P after Bedrock - they haven't enabled receipts transfer in the protocol (or at least I couldn't find one). Arbitrum is even sadder - I don't believe there is a P2P layer at all - you just have to run your own node, hold state and execute blocks to get events. There is hope - Paradigm recently released Ress - stateless execution, but it requires nodes to support Witness preparation & exchange - but this could work for L2s - cause the main blocker for local RPCs rn is huge state (VPS with TB storage cost a lot), and the second blocker is EVM forks makes it hard to hold a node - it needs to be maintained, upgraded, etc. Ress at least solves the state part. But anyways, I will try to continue working on this and release some MVP version with RPC endpoint and data storage soon - follow the updates!show more

AI just hit a wall that no amount of... money can move. The planet itself. There is not enough power, water, or land on Earth to build the data centers the AI race now demands. So the most valuable bet in artificial intelligence is no longer a chip company or a model. It is a rocket company. The plan is to leave. In January, SpaceX filed with the FCC to launch up to 1 million solar-powered data center satellites into orbit. In February it bought xAI, the maker of Grok, folding an entire frontier AI lab into a rocket company in the largest corporate merger ever recorded. On June 8 it unveiled the AI1, a compute satellite with a 70-meter wingspan, wider than a Boeing 747, powered by the sun, cooled by the vacuum of space, and wired to the ground through Starlink. Four days later it went public in the largest IPO in history, near 1.77 trillion dollars, touched 2.1 trillion on its first day, raised close to 86 billion, and made one man the first trillionaire alive. Now read the direction of that merger, because it is the whole story. A rocket company bought the AI lab. Not the reverse. For three years everyone assumed the constraint on AI was chips, or data, or talent. It is none of them anymore. It is energy and heat and dirt. The head of Anthropic said his company grew faster than the exponential, 80 times in a single year, and that is exactly why it ran out of compute. The answer was not to build more data centers in Virginia. It was to leave the atmosphere, where the sun never sets and a solar panel does five times the work. The moat in artificial intelligence is no longer the model. It is the launch. And the first rent is already being paid. A rival lab, Anthropic, is reported to be sending roughly 1.25 billion dollars a month to Musk for compute. Google near 920 million. If intelligence moves to orbit, the company that owns the only affordable road there becomes the landlord of the next layer of the internet, the way one bookstore became the landlord of the cloud. The merger is the proof of concept. The IPO is the war chest. Those monthly checks are the lease. Here is the part the price tag does not want you to read. Close to a trillion dollars of that valuation rests on orbital data centers that do not yet exist, and on a chip factory, Terafab, that SpaceX's own public filing calls a general framework with no binding deal, one that may not achieve commercial viability. Musk said it on camera. This is not a promise. The largest IPO ever written is priced on a future the filing itself cannot verify. The other side is just as real. Compute in orbit costs about four times what it costs on the ground today, and the curve may not cross for fifteen years. The machines that print the chips are backordered for years. Shedding heat in a vacuum at this scale has never been done. Musk's timelines have a long history of meaning later. And Bezos is racing the same orbit with a constellation of 51,600 satellites of his own. But strip it all away and the trade underneath is one sentence. Earth has run out of room for intelligence, and whoever owns the road off the planet owns whatever gets built next. Call it the most expensive science fiction ever sold, or the first time the map of the internet pointed up.show more

hey if you have a 3060, or any GPU... with 8GB or more sitting in a drawer right now, that thing can run 9 billion parameters of intelligence autonomously. and you don't know it yet. 2 hours ago i posted that 9B hit a ceiling. 2,699 lines across 11 files. blank screen. said the limit for autonomous multifile coding on 9 billion parameters is real. then i audited every file. found 11 bugs. exact file, exact line, exact fix. duplicate variable declarations killing the script loader. a canvas reference never connected to the DOM. enemies with no movement logic. particle systems called on the class instead of the instance. fed that list as a single prompt to the same Qwen 3.5 9B on the same RTX 3060 through Hermes Agent. it fixed all 11. surgically. patch level edits across 4 files. no rewrites. no hallucinated changes. game boots. enemies spawn, move, collide. background renders. particles fire. and here's what nobody is talking about. this is a 9 billion parameter model running a full agentic framework. Hermes Agent with 31 tools. file operations, terminal, browser, code execution. not a single tool call failed. the agent chain never broke. most people think you need 70B+ for reliable tool use. this is 9B on 12 gigs doing it clean. the model didn't fail. my prompting strategy did. the ceiling is not the parameter count. the ceiling is how you prompt it. this is not done. bullets don't fire yet. boss fights need wiring. but the screen that was black 2 hours ago now has a full game rendering in real time. iterating right now. anyone with a GPU from the last 5 years should be paying attention to what is happening right now.show more

Microsoft made 100B parameter models run on a single... CPU. bitnet.cpp: The official inference framework for 1-bit LLMs. The math behind 1-bit LLMs is what makes them revolutionary. Traditional LLMs use 16-bit floating point weights. Every parameter is a number like 0.0023847 or -1.4729. When you run inference, you multiply these floats together. Billions of times. That's why you need GPUs, they're optimized for floating point matrix multiplication. BitNet b1.58 uses ternary weights: {-1, 0, 1}. That's not a simplification. That's a fundamental change in the math. When your weights are only -1, 0, or 1: → Multiply by 1 = keep the value → Multiply by -1 = flip the sign → Multiply by 0 = skip entirely Matrix multiplication becomes addition and subtraction. No floating point operations. No GPU required. This is why bitnet.cpp achieves: → 2.37x to 6.17x speedup on x86 CPUs → 1.37x to 5.07x speedup on ARM CPUs → 71.9% to 82.2% energy reduction on x86 → 55.4% to 70.0% energy reduction on ARM The speedups scale with model size. Larger models see bigger gains because there are more operations to simplify. A 100B parameter model running at human reading speed (5-7 tokens/second) on a single CPU. That's not optimization. That's a different paradigm. Why 1.58 bits? Because log₂(3) ≈ 1.58. Three possible values = 1.58 bits of information per weight. The key insight: These models aren't quantized after training. They're trained from scratch with ternary weights. The model learns to work within the constraint. No precision loss. No quality tradeoff.show more

Introducing Pods Hyperspace Pods lets a small group of... people - a family, a startup, a few friends, to pool their laptops and desktops into one AI cluster. Everyone installs the CLI, someone creates a pod, shares an invite link, and the machines form a mesh. Models like Qwen 3.5 32B or GLM-5 Turbo that need more memory than any single laptop has get automatically sharded across the group's devices - layers split proportionally, inference pipelined through the ring. From the outside it looks like one OpenAI-compatible API endpoint with a pk_* key that drops straight into your AI tools and products. No configuration beyond pasting the key and changing the base URL. A team of five paying for cloud AI burns $500–2,000 a month on API calls. The same team's existing machines can serve Qwen 3.5 (competitive on SWE-bench) and GLM-5 Turbo (#1 on BrowseComp for tool-calling and web research) for free - the hardware is already on their desks. When a query genuinely needs a frontier model nobody has locally, the pod falls back to cloud at wholesale rates from a shared treasury. But for the daily work - code reviews, refactors, research, drafting - local models handle it and nobody gets billed. And when it is idle, you can rent out your pod on the compute marketplace, with fine-grained permissions for access management. There's no central server involved in inference. Prompts go from your machine to your pod members' machines and back: all of this enabled by the fully peer-to-peer Hyperspace network. Pod state - who's a member, which API keys are valid, how much treasury is left - is replicated across members with consensus, so the whole thing works on a local network. Members behind home routers don't need port forwarding either. The practical setup for most pods is three models covering different jobs: Qwen 3.5 32B for code and reasoning, GLM-5 Turbo for browsing and research, Gemma 4 for fast lightweight tasks. All running on hardware you already own. Pods ship today in Hyperspace v5.19. Model sharding, API keys, treasury, and Raft coordinator are all live. What Makes This Different - No middleman. Your prompts travel from your IDE to your pod members' hardware and back. There is no server in between reading your data. - No vendor lock-in. Pod membership, API keys, and treasury are replicated across your own machines using Raft consensus. If the internet goes down, your local network keeps working. There is no database in someone else's cloud that your pod depends on. - Automatic sharding. You don't configure layer ranges or calculate VRAM budgets. Tell the pod which model you want. It figures out how to split it across whatever hardware is online. - Real NAT traversal. Your friend behind a home router with a dynamic IP? Works. No VPN, no Tailscale, no port forwarding. The nodes handle it. - Free when local. This is the part that matters most. Cloud AI bills scale with usage. Pod inference on local hardware scales with nothing. The marginal cost of your 10,000th prompt is the electricity your laptop was already using. Coming soon: - Pod federation: pods form alliances with other pods. - Marketplace: pods with spare capacity can sell inference to other pods.show more

Stanford researchers did it again. They just built the... agent-native version of Git. When an agent works on a longer task, the run builds up a lot of state. This includes files edited/created, a dev server, a database, installed packages, KV cache, etc. Say the agent is at step 10 and makes a mistake, maybe it misreads a traceback and rewrites a file that was actually fine. The tests start failing, and the run goes off track, although everything through step eight was correct. By default, the agent just tries to fix it, which creates more edits and tool calls. This burns more tokens and grows the context. The other options are a person stepping in to redirect it or restarting the whole run from step one. That's wasteful, because it pays for every model/tool call again and re-prefills the context. Moreover, since an agent's run is non-deterministic, it doesn't reproduce the same early steps anyway. The reason it's hard to just jump back exactly to a previous correct step and resume from there is that the trajectory is only a message log. It records what the agent said and which tools it called, but not the live state underneath. That state includes things like memory, open file handles, child processes, installed packages, /tmp, and KV cache. None of that is in the log. Git can version the files, but it doesn't snapshot the running process or the KV cache. Checking out step eight moves the files back, but the process is still sitting in step-ten memory with a cold cache. Shepherd is a runtime layer by Stanford that records the run as a trace of typed events rather than a flat log. Each agent-environment interaction becomes a commit, similar to Git, but it tracks the live run. Its commit includes the agent process and the filesystem together, copy-on-write, so a branch carries the actual state and not just the files. Going back to a previous step is then a single call that forks from that commit and continues from the exact state. The copy-on-write fork is roughly five times faster than docker commit, and because the prompt prefix through step eight is unchanged, the KV cache is reused over 95% on replay, so early steps aren't reprocessed again. Once the run can be forked, a meta-agent can sit on top and operate it. It watches the trace and reverts as soon as it looks wrong, before the bad write is committed. In practice, it's just Python calling fork, replay, and revert on the trace, rather than a separate control plane wired into the harness. Not everything is reversible though. Files and sandbox changes undo themselves, but a database write has no automatic undo, so it needs a matching undo step set up in advance. Something external, like a sent email or a real charge, can't be undone, so the supervisor's job there is to catch it before it fires. They tested this on a few public benchmarks. On CooperBench, where two agents work on the same codebase, adding a live supervisor took the pair-coding pass rate from 28.8% to 54.7%. It's still early and labeled alpha. The benefit mostly shows up when a run gets branched a lot over a heavy sandbox state, which is exactly where restarting wastes the most tokens and time. If Git was made to make file changes reversible, Shepherd is trying to do the same thing for a live agent run. Shepherd Repo: (don't forget to star it ⭐ ) That said, Shepherd reverts a bad step inside a run. The harness around it, the prompts, tools, and checks the supervisor relies on, still drifts across runs as models and dependencies change. Akshay wrote about making that harness repair itself, where a failing trace gets diagnosed, the fix is verified against the exact input that failed, and the failure is locked as a regression test so it can't recur. Read it below.show more

introducing a new, very fun, LLM benchmark- the Game-of-Life... Bench! the rules are simple: given an 8x8 grid following Conway's game of life rules, the goal is to create an initial pattern with at most 32 cells that can last the longest number of turns before dying/repeating. some results to highlight (with caveats detailed below): - gpt 5.1 lasts the longest with a 106 step run - claude models are really bad at this! they refuse to reason about this task and score < 25 points - deepseek r1 is the best open model with 102 steps. why? because i wanted to create a benchmark that has (i think) no practicality, but is still fun to look at, cheap, and still measures something interesting. i also am a big fan of the game of life. its absurdly simple rules leading to intractability is extremely cool to me. also, i saw a lot of work with LLMs trying to "predict" the next state in Conway's game of life, I think game-of-life bench is more fun because it's pretty open ended and only asks the LLM for the initial state. I also think this could be an RL env? but idk why you would ever train on this task haha i don't think this is a "serious" benchmark because it doesnt measure anything practical, but i still think it's a hard benchmark exactly because you can't predict what happens with your initial state many turns into the future; this is why i was initially expecting all LLMs to be bad at it, but turns out, some are clearly better than the others (the ordering may surprise you!) reminder: this is still a work-in-progress; (1) i am gpu-poor so could only do 10 runs for each model, even though total running cost is relatively low. maybe with some more credits i can run more seeds for each model. (2) i handpicked models which i think are at the frontier right now, plus some others that were on my mind. so, if you'd like to see a model on here, let me know. (3) i currently only do an 8x8 grid because i thought that by itself would be pretty hard for current LLMs, but of course we can increase grid sizes! (4) the coolest thing is, i dont think we can calculate the max possible number of states (yay undecidability!) you can go without repeating, so this is essentially a no-ceiling task, which is pretty cool! again, i did this mostly out of a desire to make LLMs do something fun. if this keeps me entertained for a few more days, i'd likely release a blog post on it. if it keeps me entertained for a week (and someone sponsors me), i'll put more work into it :P lastly, this is fully open sourced, so feel free to run this on your own!show more