Nvidia's Risky Business
• Jay Cooke's 1870 financing for Northern Pacific Railway led to the 1873 Panic and economic depression.Read more | Comments | @hackernws
• In 2026, tech companies Microsoft, Oracle, Meta, Alphabet, and Amazon have issued $194 billion in debt for AI infrastructure.
• Google raised $85 billion equity in June 2026, including $10 billion from Berkshire Hathaway, to fund AI build-out.
• DeepMind faces leadership changes and loss of frontier AI status with Google Cloud gaining more compute resources.
• Nvidia partners with major financiers to mobilize $500 billion for AI infrastructure as an investable asset class, risking 25% residual-value financing.#AIInfrastructure #Nvidia #TechFinance
Stratechery by Ben Thompson
Nvidia’s Risky Business
Nvidia is finding new ways for its customers to raise money, and it’s expanding the risk of the AI buildout significantly.
ATProto for Distributed Systems Engineers
• AT Protocol uses a NoSQL cluster with eventual consistency for scalability.Read more | Comments | @hackernws
• Introduces an event log (e.g., Kafka) to sync view servers with the NoSQL cluster.
• Data is organized in user data repositories with collections of JSON document records.
• Every internal service is externalized with public APIs for decentralized backend participation.
• Applications read/write JSON repos, emit events to logs, and update view servers in a circular data flow.#ATProtocol #DistributedSystems #Decentralization
AT Protocol
Atproto for distributed systems engineers - AT Protocol
In this article we're going to explore atproto from the perspective of distributed backend engineering.
Apple Silicon and macOS VMs: 11–16× Faster LLM Inference with Llama.cpp
• A Metal capability shim boosts LLM inference in macOS VMs up to 16x faster on M1 Ultra.Read more | Comments | @hackernws
• Shim changes Apple GPU capability reports, enabling newer Metal kernels in llama.cpp.
• TinyLlama 1.1B prompt processing was 11.08x faster; token generation 16.36x faster inside VM.
• Tested on macOS 26.6.1, shim achieves near bare-metal performance (up to 99.59%) for some models.
• Shim is experimental, per-process, uses private Metal APIs, and limited to current Apple Silicon and guest versions.#AppleSilicon #GPUAcceleration #MacOSVM
GitHub
cua/blog/gpu-passthrough-macos-vms.md at main · trycua/cua
Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation. - trycua/cua
LFM2.5 2.6B model competitive with 4x larger models
• LFM2.5-2.6B is a 2.69B parameter hybrid model with a 128K context window for on-device deployment.Read more | Comments | @hackernws
• It achieves 220 tokens/s inference on Apple M5 Max and 113 tokens/s on AMD Ryzen CPU using under 2.5 GB RAM.
• Trained on 34 trillion tokens with agentic reinforcement learning for tool use and multi-step tasks.
• Available in native, GGUF, ONNX, and MLX formats optimized for various hardware including Apple Silicon.
• Benchmark tests show it outperforms larger models (4x size) in tool use and multi-agent tasks.#LiquidAI #OnDeviceAI #LFM2Model
huggingface.co
LiquidAI/LFM2.5-2.6B · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Woman Pulled over at Gunpoint Twice After Flock Camera Glitch
• Amber Newell of Brookfield, Wisconsin, was pulled over twice at gunpoint due to a Flock camera glitch.Read more | Comments | @hackernws
• The camera falsely flagged her car in connection to a homicide investigation on I-94 near Milwaukee.
• Milwaukee police admitted the alert was outdated and should have been removed days earlier.
• Brookfield officers followed protocol given the alert, despite the mistake being a human error.
• Newell's daughter is now fearful of police stops, highlighting the impact of the error.#FlockCameraGlitch #FalsePoliceStop #PoliceTechnology
Guessing Headlights
"Y'all Failed Me": Woman Pulled Over at Gunpoint Twice After Flock Camera Glitch
Technology is supposed to make police work faster and safer. Most of the time it does. But every so often, a computer system hiccups, and an ordinary person
Go is an ideal language for AI-assisted software engineering
• Go language focuses on software engineering with strong team-driven development principles.Read more | Comments | @hackernws
• Go is a platform with built-in tools: formatter, test framework, dependency management, security.
• Go enforces readability and uniform code format, aiding AI and human code review.
• Go has a strong static type system and comprehensive standard library for secure, reliable code.
• Go guarantees backward compatibility and provides tools for large-scale, maintainable codebases.#GoLang #AIcoding #SoftwareEngineering
Googleblog
Google for Developers Blog - News about Web, Mobile, AI and Cloud
As AI shifts software engineering from writing to reviewing, discover how Go's strict compiler and unified toolchain ensure reliable AI-generated code.
Mojo 1.0
• Mojo language reaches stable 1.0 release after initial 2023 launch.Read more | Comments | @hackernws
• Mojo 1.0 provides stable foundation for long-term development, minimizing breaking changes.
• New features include Python-style lambda syntax and improved memory safety diagnostics.
• MAX framework updated with easier installation and support for GLM-5.2 and Nemotron-H models.
• Modular plans to open source Mojo compiler and toolchain within 2026.
• Upcoming ModCon event on August 18 will discuss Mojo 1.0 and open source plans.#MojoLang #ModularAI #AIProgramming
Modular
Modular: Modular 26.5: Mojo 1.0 is here!
Today, the Mojo language officially reaches 1.0: a milestone the language has been building toward since its first release in 2023. Mojo has grown into a general-purpose language with a vibrant developer community writing their own libraries, tools, and applications…
OpenAI’s head of ethics leaves less than a year after joining
• OpenAI's head of ethics left the start-up less than one year after joining.Read more | Comments | @hackernws
• The departure occurred before August 10, 2026, as reported by Financial Times.
• No specific reasons for the departure were disclosed in the article.
• The Oxford-based start-up had recently hired the ethics head from OpenAI.
• The exit highlights challenges in retaining top ethics executives in AI start-ups.#OpenAI #AIethics #TechLeadership
London Underground begins scanning passengers' faces
• BTP is expanding its live facial recognition (LFR) trial into London Underground stations.Read more | Comments | @hackernws
• The trial aims to enhance public safety and crime detection on the network.
• LFR technology matches live images against a watchlist of individuals of interest.
• The expansion follows initial trials on certain rail services and stations.
• Public awareness and privacy considerations remain important in the ongoing trial.#FacialRecognition #LondonUnderground #PublicSafety
www.btp.police.uk
BTP expands Live Facial Recognition (LFR) trial into London Underground stations
British Transport Police (BTP) is to expand its trial of Live Facial Recognition (LFR) technology by deploying it in Transport for London (TfL) Underground stations.
Show HN: iPhone app takes simultaneous images from 2 lenses, fuses into 1 photo
• Photosynthesis app for iPhone combines two lenses to create one high-res photo.Read more | Comments | @hackernws
• Supports all iPhones with 2 or 3 rear lenses from iPhone 11 onwards.
• No generative AI used; detail comes solely from optical camera data.
• Offers free unlimited capture/editing, 3 exports/month; subscription for full features.
• Exports include layered Photoshop, spatial 3D, and re-importable Photosynthesis files.#iPhonePhotography #CameraApp #MobilePhotography
photosynthesis.camera
Zoom and enhance. For real this time. Two lenses, one press, one high-res photo.
Compression is prediction
• Modern data compression uses transforms, models, and entropy coders like arithmetic coding to shrink data.Read more | Comments | @hackernws
• Arithmetic coding compresses entire datasets into a single number based on symbol probabilities.
• Higher symbol probabilities allow compression to fewer bits per symbol, measured by entropy.
• Context-aware models (order-N) vastly improve compression by predicting next symbols more accurately.
• Large language models (LLMs) excel at prediction, enabling superior compression but with high computational cost.#DataCompression #LanguageModeling #EntropyCoding
ngrok blog
Compression is prediction | ngrok blog
Compression and LLMs are trying to solve the exact same problem: predicting what comes next. Learn the fundamentals of compression and how better prediction leads to better shrinkage.
What I learned by putting GitHub Copilot behind a MitM proxy
• GitHub Copilot uses an OAuth flow for authentication and entitlements validation at startup.Read more | Comments | @hackernws
• Copilot sends network requests via Electron's Chromium-based network stack, interceptable by mitmproxy.
• Copilot selects models dynamically based on intent classification like code-gen or debugging.
• Session history including user prompts and AI responses is stored in plain text SQLite database locally.
• Default context includes recent edits from up to 20 files and 3 lines around changes; no default .env exclusion on individual plans.#GitHubCopilot #AIcoding #ElectronApps
Lighthousenewsletter
I put GitHub Copilot Behind a MITM Proxy. Here's What I found.
A look inside Copilot’s network traffic, harness, memory, and how context is becoming the product.
Nvidia Nemotron 3.5 Lightning and NeMo Switchyard
• NVIDIA released Nemotron 3.5 Lightning, a 30-billion-parameter open model for specialized AI tasks, with up to 4x faster output speed.Read more | Comments | @hackernws
• Nemotron 3.5 Lightning is customizable, supports on-premise and cloud deployment, and improves accuracy in domain-specific workflows.
• NVIDIA introduced open source NeMo Switchyard, a smart routing library for directing AI requests to the best model automatically.
• NeMo Switchyard reduces task cost to about one-third while maintaining frontier-level accuracy.
• Key industry partners like CrowdStrike and Siemens have adopted Nemotron 3.5 Lightning and NeMo Switchyard for AI efficiency improvements.#NVIDIA #AIModels #AgenticAI
NVIDIA Blog
NVIDIA Nemotron 3.5 Lightning and NeMo Switchyard Deliver Faster, Smarter, More Efficient Agentic AI
The new lightweight open model and routing library delivers greater control over AI, data and workflows across edge devices, PCs, workstations, data centers and the cloud.
Jolt: Clojure compiler implemented with Chez Scheme
• Jolt is a self-hosted Clojure implementation compiling to Scheme, running on Chez and Gambit.Read more | Comments | @hackernws
• It supports full Clojure semantics including persistent data, lazy seqs, multimethods, and core.async.
• The 'jolt build' command creates standalone binaries needing no JVM or Scheme runtime.
• Real concurrency is provided via OS threads with 'future', 'agent', 'pmap', and 'core.async'.
• No general Java interop, BigDecimal absent, and regex differs using irregex instead of Java's engine.#Clojure #Scheme #JoltLang
jolt-lang.github.io
Jolt — Clojure on Scheme
Jolt is a Clojure implementation on Scheme — Chez natively, Gambit for JavaScript. Self-hosted compiler, Clojure-compatible standard library, no JVM.
Grok Bot
• Grok Bot performs tasks autonomously on desktop or iOS using your apps.Read more | Comments | @hackernws
• Users can manage multiple Bots working in parallel on different projects.
• Bots learn workflows by observing user actions and improve over time.
• Grok Bot plans to integrate collaborative task handoffs between Bots.
• Pricing: Cursor Ultra at $200/month, Premium Teams at $120/seat/month with team features.#GrokBot #AIAutomation #ProductivityTools
SpaceXAI
Grok Bot: A new kind of colleague
Grok Bot is your team of always-on AI agents that finish the work. They have their own computer, use it like you do, and never log off.
WorldClaw Agentic 3D open-world generation at scale
• WorldClaw generates large-scale, editable 3D open-world scenes from single text prompts.Read more | Comments | @hackernws
• It uses agentic planning to maintain global terrain coherence and rich local detail.
• The system produces explicit terrains plus independently editable 3D objects with accurate placement.
• WorldClaw's pipeline includes intent analysis, global terrain generation, and regional object placement.
• Future plans involve code-native 3D modeling and integration with game engines for runtime use.#3DGeneration #AIModels #WorldClaw
WorldClaw
WorldClaw: Agentic 3D Open-World Generation at Scale
From one open-ended prompt to an explicit, explorable, and editable 3D world.
Manus will return to operating as an independent company
• Manus will become independent and delete some users' data from Aug 23-24, 2026 (SGT).Read more | Comments | @hackernws
• Affected users can back up data until Aug 23, 2026, 7:59 a.m. SGT and restore from Aug 25, 2026.
• Unaffected users can use Manus normally with no action required or charges during backup period.
• Notifications will be via app and email; users registering via Apple ID or Facebook should check app notifications.
• Data deletion affects data generated on/after Dec 29, 2025, due to regulatory compliance in some jurisdictions.#ManusUpdate #DataBackup #TechTransition
manus.im
A Note to Our Users
Visit the Manus blog to learn more about our thoughts on the future.
Show HN: Git-knife – Edit commit messages, authors, and dates like a spreadsheet
• git-knife is a desktop GUI tool for editing git commit metadata including message, author, and dates.Read more | Comments | @hackernws
• It enables bulk find & replace with regex and edits author/committer names, emails, and dates.
• git-knife shells out to git CLI and rebuilds commits with git commit-tree, ensuring file contents remain unchanged.
• It supports Linux, macOS, and Windows builds, requiring git, Node.js, pnpm, and Rust as prerequisites.
• The tool creates automatic backups before rewrites and warns when editing pushed history, but does not yet handle reorder, squash, or merge commits.#GitTools #OpenSource #DeveloperTools
GitHub
GitHub - TheRealYT/git-knife
Contribute to TheRealYT/git-knife development by creating an account on GitHub.
CFTC declares market emergency, orders Kalshi to continue to operate in New York
• CFTC used emergency authority to keep KalshiEX operating under federal Commodity Exchange Act.Read more | Comments | @hackernws
• KalshiEX's market emergency arose after NY Attorney General sued, seeking $36B damages and halt.
• NY lawsuit aims to block KalshiEX’s event contracts citing state gaming laws.
• CFTC emphasized federal jurisdiction over derivatives exchanges, rejecting state gaming regulation.
• CFTC has filed multiple lawsuits against states to protect federal oversight of derivatives markets.#CFTC #KalshiEX #Derivatives
CFTC
CFTC Exercises Emergency Authority to Ensure Market Stability
The CFTC exercised its emergency authority in response to KalshiEX, LLC’s notification of a market emergency and ordered the exchange to continue to operate in accordance with the Commodity Exchange Act’s Core Principles.
Making holograms with a pen plotter
• Pen plotter repurposed to etch holograms by creating reflective ridges encoding depth.Read more | Comments | @hackernws
• Reflective ridge curvature controls highlight movement, simulating 3D depth perception.
• CD jewel cases found effective medium for etching holographic patterns with a sharp stylus.
• Point-source flashlight required for best holographic illusion due to light reflection properties.
• Etched holograms retain complete scene info even when cut into small pieces, emphasizing holographic nature.#Holography #PenPlotter #DIYTech
Matelsky
Making holograms with a pen plotter
I have a pen-plotter!
It’s great.
It’s like having a printer but slow and it breaks more often.
I got it on eBay and it came sorta broken and I immediately spent the difference between what I paid and the price of new one fixing it.
Now it’s fixed!
It’s great.
It’s like having a printer but slow and it breaks more often.
I got it on eBay and it came sorta broken and I immediately spent the difference between what I paid and the price of new one fixing it.
Now it’s fixed!
New Bedford police officer accused of using Flock cameras to track ex-partner
• New Bedford officer Emily Pacheco accused of tracking ex-girlfriend using Flock license plate readers.Read more | Comments | @hackernws
• Judge ordered Pacheco to surrender firearms and suspended her carry license after abuse prevention order.
• Police department suspended Flock camera use and opened internal investigations, placing an officer on leave.
• Pacheco performed hundreds of vehicle searches from April to June, often citing 'traffic infraction' as reason.
• Public and city councilors demand regulation or removal of Flock cameras amid privacy abuse concerns.#PoliceSurveillance #PrivacyAbuse #NewBedford
The New Bedford Light
New Bedford police officer accused of using Flock cameras to track and follow ex-romantic partner
Officer Emily Pacheco faces an internal affairs investigation. She was served an abuse prevention order Monday after a judge’s ruling.