Today’s biggest story in AI infrastructure is the successful deployment of deep learning language models on AMD’s MI300X GPU. Ryan Zhou shared a production-ready configuration to run DeepSeek V4 Flash on a single MI300X with 156.67 GiB HBM3 memory, outperforming Nvidia’s H100 in capacity and cost, and enabling up to 64 concurrent streams without errors. Apple’s legal battle with OpenAI intensified as OpenAI publicly countered Apple’s trade secret claims, sharing communications that dispute Apple’s accusations involving ex-Apple employees. Spotify announced surpassing 300 million subscribers despite price hikes and backlash against AI-generated music, while Runware launched portable modular AI inference data center pods aimed at flexible and scalable AI model deployment worldwide. Additionally, a massive npm supply chain attack compromised dozens of widely used caching libraries, distributing a credential-stealing worm with over 2 billion monthly installs affected.
AI & Machine Learning
- DeepSeek V4 Flash on a Single AMD MI300X: Ryan Zhou released a stable setup to run a 304B-parameter AI model on one AMD MI300X GPU using full HBM3 memory, demonstrating higher capacity and lower cost than Nvidia’s top hardware. This enables efficient handling of multiple inference streams without errors.
- LLMs reward expertise: A deep dive into how domain expertise remains critical to getting valuable results from large language models, exemplified by Terence Tao’s superior mathematical prompting compared to general users.
- Harness Engineering for Self-Improvement: Discussion on the role of harness architectures in enabling recursive self-improvement for AI, focusing on systems that orchestrate model execution, planning, memory, and tool use.
- Show HN: Fine-tune an 8B model on a 4 GB laptop GPU: Soup is a simplified CLI tool for fine-tuning large language models with minimal configuration, supporting techniques like QLoRA directly on consumer GPUs without requiring the cloud.
- Show HN: Run an 80B Qwen in 4.3 GB of RAM on a Mac, and a 35B on an iPhone: Swiftlet demonstrates running large mixture-of-experts LLMs with runtime streaming on Apple devices, including iPhones, enabling massive models to operate under tight memory constraints.
- Why Large Language Models Fail at Tabular Prediction: Research revealing that LLMs struggle with high-dimensional tabular data, which limits their performance for predictive analytics relative to classical machine learning methods.
- Ten advances in mathematics and theoretical computer science: OpenAI revealed results from an internal model (Astra) that contributed to significant progress on longstanding mathematical problems, including a disproof of the Erdős unit-distance conjecture.
- Mistral Is in the Right Place at the Right Time: Mistral, a European open-source AI lab, capitalizes on turbulence in US AI labs and regulatory restrictions to gain traction and funding as the industry questions proprietary closed-weight models.
- Design Arena creators raise $7.9 million to bring taste to AI models: Design Arena, a startup using scalable human feedback to improve AI-generated games and media, secured seed funding to expand its service as a quality evaluation platform for frontier labs.
Security & Privacy
- Keyv and friends compromised in active Shai-Hulud supply chain attack: Attackers hijacked the maintainer’s GitHub account to inject a credential-stealing worm into widely used npm caching libraries, impacting over 868 packages with 2 billion+ monthly downloads and spreading rapidly.
- Microsoft: Suspected Russian Hackers Are Targeting Logins Via Hotel Wi-Fi: Microsoft warned of Russian cyberattacks hijacking hotel Wi-Fi to steal Microsoft 365 credentials and distribute trojans disguised as legitimate software updates.
- Apple challenges UK government’s latest demand for iCloud backdoor: report: Apple filed a legal complaint against the UK government’s secret order demanding access to encrypted iCloud data, continuing an ongoing privacy battle.
- Apple briefly yanked Telegram from the App Store over CSAM violations: Apple temporarily removed Telegram due to child sexual abuse material but restored it after Telegram took swift action to remove offending content and ban users.
Business & Industry
- Spotify now has over 300M subscribers: Spotify announced surpassing 300 million subscribers and 777 million MAUs while expanding AI-driven personalized content and experimenting with AI-powered podcast features.
- Bending Spoons to buy Airtable for $1.28B: Bending Spoons, newly public, is acquiring Airtable for $1.28 billion cash, seeking to accelerate innovation and streamline the company’s operations amid valuation drops.
- After killer quarter, Palantir CEO Alex Karp calls AI industry ‘Marxist’: Palantir CEO criticized AI frontier labs for centralizing power, despite Palantir's own record revenue driven by AI adoption in enterprise.
- Snap CEO sidesteps Specs preorder questions on Q2 earnings call: Snap’s CEO avoided detailed comments on preorder interest for its $2,195 Specs smart glasses, emphasizing the technical complexity and upcoming product launch.
- AWS is helping vibe-coding startup Superblocks, and the implications are big: AWS formed a marketing and technology partnership with Superblocks to enable enterprise vibe coding tools inside private AWS clouds with full data control.
- OpenAI drags Apple’s lawsuit into the court of public opinion: OpenAI publicly refuted Apple’s trade secret theft lawsuit claims with leaked message transcripts, trying to shape perception ahead of legal proceedings.
Hardware & Infrastructure
- Runware builds a pod to find out if the future of data centers is portable: Runware unveiled a modular, transportable AI inference data center pod designed to offer scalable, flexible, and energy-efficient AI compute to multiple regions easily.
- EON wants to move the data superhighway from ocean fiber to space lasers: Startup EON launches with $10.75M seed funding aiming to establish a laser satellite network capable of 2.4Tbps links between data centers, challenging undersea fiber optics.
- SpaceX is set to acquire 130,000 acres of marshland in southern Louisiana: SpaceX nears closing on a massive coastal land parcel for future Starship launch operations, expanding its US launch infrastructure.
- Here's How AMD Zen 6 Could Help Eliminate Gaming Frame Rate Stutters: AMD’s upcoming Zen 6 CPU architecture will introduce core prioritization features improving gaming smoothness by reducing frame dips through smarter frequency management.
- Samsung’s discounted smart monitor is $349.99, its lowest price yet: Amazon and B&H discounted Samsung’s 32-inch 4K M80F smart monitor, a versatile display with video streaming apps, webcam, and cloud gaming support.
Entertainment & Culture
- Ray Bradbury's "There Will Come Soft Rains" is set today (2026-08-04): On this day in 2026, we revisit Bradbury’s prophetic story about an automated house continuing its functions long after its inhabitants are gone.
- The studio behind Pokémon struggles to find its voice in Beast of Reincarnation: Game Freak’s new action game experiment shows polish but lacks originality, according to a review.
- Big Walk, Dying Light 2, and Signalis Hit PlayStation Plus Tomorrow: Sony adds three games including Big Walk to PlayStation Plus, raising questions about the value of paid game subscriptions as physical media declines.
- The Best Gaming Mouse You Can Buy After Testing Dozens of Models: WIRED updates its gaming mouse buyer’s guide with new recommendations including the Razer Viper V4 Pro.
- Is This Poker Player Bluffing? The AI Thinks So: ESPN uses AI to analyze poker player tells during the 2026 World Series of Poker, generating mixed reactions from the community.
Other Tech & Culture
- Xbox goes down. You can't play games you own on disc: A multi-day Xbox outage demonstrated that even physical disc owners could not play games, highlighting the increasingly digital-dependent nature of modern consoles.
- WhatsApp says it is fixing an issue that disabled several accounts: Meta works to restore access after numerous WhatsApp users were temporarily locked out and put “under review.”
- Europe’s AI labeling and transparency rules are now in effect: The EU’s AI Act transparency requirements went live, mandating AI-generated content disclosure and labeling with potential fines for non-compliance.
- Samsung Blocks TV Apps That Let Others Use Your Internet Connection: Samsung bans smart TV apps that exploit users’ IP addresses for residential proxy services after a cybersecurity investigation.
- Your Fitbit data can now connect directly to Apple Health: Google Health added direct syncing of Fitbit data to Apple Health, improving cross-platform health data integration.
- India moves to give its instant payments network a business model: India legislates to potentially introduce merchant fees on UPI payments, changing the previously free-to-merchant payments network for sustainability.
- Trump’s AI protectionism has come for robotics: The FTC enacted bans on imported foreign robots, citing national security risks and to protect US robotics firms amid an AI industrial policy push.