AMD's AI Future Is Rack Scale 'Helios' (morethanmoore.substack.com | comments) AMD has announced the launch of 'Helios', a rack-scale AI system designed to enhance computational efficiency for complex AI tasks and applications. This initiative marks a significant step in advancing AMD's capabilities in artificial intelligence and aims to position the company as a leader in AI solutions for 2025.

• If hope AMD can produce a chip that matches H100 in training workloads.
• AMD future should be figuring out how to reproduce the performance numbers they “claim” they are getting

Waymo rides cost more than Uber or Lyft and people are paying anyway (techcrunch.com | comments) A new analysis from ride-hailing aggregator Obi reveals that rides from Waymo, the autonomous taxi service, cost more than those from Uber or Lyft, especially for shorter trips, where passengers experience longer wait times. Despite higher costs, consumers continue to choose Waymo’s services.

• A robot isn’t going to decide it doesn’t want to take my ride after accepting it and drive around aimlessly hoping I’ll get tired of waiting and cancel. I haven’t needed Uber/Lyft on a regular basis in several years, but back when I did that was a frequently recurring problem.

Saab achieves AI milestone with Gripen E (saab.com | comments) Saab, in partnership with Helsing, has completed three test flights integrating the AI agent 'Centaur' into its Gripen E fighter jet. This achievement enhances the aircraft's operational capabilities.

• What about blockchain though? Does it have that?
• What kind of models do these run? Vision? What “predictions” are being made?
• They should replace some of the panels with machined metamaterial and give it some RF absorbing paint. Would sell like hotcakes in the current political climate.

Google Cloud Incident Report – 2025-06-13 (cloud.google.com | comments) On June 13, 2025, Google Cloud reported multiple service issues affecting various Google Cloud Platform (GCP) products. The status report includes information for users about service availability, disruption, and outages, encouraging users to check current service status or contact support if they encounter unlisted issues.

• > Without the appropriate error handling, the null pointer caused the binary to crash. We must be at the trillion dollar mistake by now, right?
• Usually Google and FAANG outages in general are due to things that happens only at Google scale but this incident seems from a generic small/medium company with 30 engineers at most.

The Many Sides of Erik Satie (mitpress.mit.edu | comments) Erik Satie's life and music present a complex tapestry that intertwines clarity with intrigue. His compositions, famously paradoxical, are celebrated for their freshness and profound elegance, contrasting with the murkier elements of his personal life. His innovative approach has influenced various musical genres and continues to resonate with audiences today.

• Hate to nit, but there is actually a seventh gnossienne. Satie didn't publish 4-6, or even label them "gnossiennes", whereas the seventh was explicitly referred to as a gnossienne by Satie.
• His music appears in the soundtrack for the beautiful comedy movie Being There, with Peter Sellers, along with some lovely matching pieces by Stephen Edwards.

Wrong ways to use the databases, when the pendulum swung too far (luu.io | comments) The narrative recounts a junior developer's experiences with a critical database system fraught with performance issues, stemming from an over-reliance on stored procedures and a resultant rewrite attempting to simplify architecture by using key-value stores. This transition created its own challenges, including inefficient data handling and increased latency due to a lack of atomic operations and transaction management.

• This was a good read, especially for me considering I’m always having to remind people that SQL has been around since the 80s and has been optimized for a number of tasks over a very long time, tasks that should live in the database vs in code Very nice to see the other side of this

Cray versus Raspberry Pi (aardvark.co.nz | comments) The author contrasts the Cray 1 supercomputer, an icon of 1970s computing with its 160 MFLOPS performance and high price tag, against the modern Raspberry Pi 5 (RPi5), which delivers up to 30 GFLOPS at a fraction of the cost and power consumption. The piece reflects on the evolution of computing hardware over the past few decades.

• > but then again if you'd showed me an RPi5 back in 1977 I would have said "nah, impossible" so who knows? I was reading lots of scifi in 1977, so I may have tried to talk to the pi like Scotty trying to talk to the mouse in Star Trek IV. And since you can run an LLM and text to speech on an RPi5, it might have answered.

Debunking HDR [video] (yedlin.net | comments) A video titled 'Debunking HDR' by Steve Yedlin addresses common misconceptions about High Dynamic Range (HDR) technology in visuals. The content includes sections on color spaces and conversion issues between Standard Dynamic Range (SDR) and HDR, outlining how perception impacts tonal representations and practical challenges for filmmakers.

• I read a whole book about this last year and it made me furious. Well, technically the book was about ACES but it was also about HDR and the problems overlap tremendously. I emphatically agree with this entire video and I will be sharing it widely.

Seven replies to the viral Apple reasoning paper and why they fall short (garymarcus.substack.com | comments) The discussion critiques a recent Apple reasoning paper, addressing seven specific responses that lack critical points, diminishing their effectiveness in challenging the original paper's claims. This highlights deeper issues within the debate on the topic.

• Why do we keep posting stuff from Gary? He's been wrong for decades but he keeps writing this stuff. As far as I can tell he's the person that people reach for when they want to justify their beliefs. But surely being this wrong for this wrong should eventually lead to losing ones status as an expert.

Me an' Algernon – grappling with (temporary) cognitive decline (tidyfirst.substack.com | comments) In 'Me an' Algernon', Kent Beck reflects on his experiences with cognitive decline, drawing parallels to the story of 'Flowers for Algernon'. He explores the emotional and practical challenges of temporarily losing mental clarity, while considering the societal impacts of cognitive changes. Beck's insights touch on fear, identity, and the value placed on intelligence.

• What is “the genie” that is referred to a few times? An AI coding copilot?
• If you experience temporary cognitive decline like this definitely pay attention to it, it can indicate serious stressors in your life that need to be dealt with immediately. It can get worse steadily if you ignore it!

"The Illusion of Thinking" – Thoughts on This Important Paper (learningbyshipping.com | comments) A paper argues that AI should not be anthropomorphized, as this misconception hinders its development. It draws a line from historical examples like Clippy to current large language models (LLMs) and chatbots, suggesting this viewpoint is crucial for understanding AI capabilities. The author expresses strong appreciation for the paper's insights.

• The conversational interface is a bit of a dark pattern, too, because it exaggerates what an LLM can do and creates an illusion that makes using LLMs a bit more addictive than a technical interface that says "generating random data that may be useful".

How to Bring Back Oddly Shaped App Icons in macOS 26 Tahoe (simonbs.dev | comments) macOS 26 Tahoe replaces unique app icons with uniform squircle icons, sacrificing the distinct personality of the Mac. Users can restore custom icon shapes through Finder, while developers can utilize the NSDockTile API to display unique designs. Step-by-step instructions are provided for both users and developers.

• Blows my mind how slow OSX still is, there is perceptible lag in almost every interaction during the gif in the blog post. Opening the "properties" dialog looks like it takes > 1 second. It's 2025, we have amazing processors, why is this software still so bad.

TimeGuessr (timeguessr.com | comments) TimeGuessr is an interactive game that challenges players' knowledge of geography and history by asking them to identify the locations and dates where historical photos were taken. The platform offers features like daily challenges and user accounts, creating an engaging learning experience.

• This is fun! I didn't immediately notice the time slider and thought it was just GeoGuessr with historical pictures, but the time element adds a whole new dimension.
• I couldn’t figure out how the time picker worked (and that there was a time picker) till the third round when I realised I was being penalized. Got 34k still.

Have a damaged painting? Restore it in just hours with an AI-generated “mask” (news.mit.edu | comments) A new restoration technique harnesses artificial intelligence to repair damaged paintings in hours, employing a removable digital 'mask' made of thin film that adheres to the artwork. This method significantly accelerates traditional restoration processes and enhances the precision of repairs.

• The restoration method is more interesting than the AI, but claiming it uses AI is probably necessary to get funding for it.
• Why not just generate a new painting with AI? How much of the painting has to be covered with a mask before it’s not a painting anymore?

Infinite Grid of Resistors (mathpages.com | comments) The puzzle of an infinite grid of resistors involves analyzing the effective resistance between adjacent nodes connected by a resistance R. Through symmetry and flow field analysis, it's concluded that this resistance equals R/2. The reasoning relies on some contentious assumptions about current flow at infinity, which complicates its proof.

• Re: the infinite resistor grid If you take an endless grid made of identical resistors and try to measure the resistance between two neighbouring points, the answer turns out to be about one-third of a resistor
• This was the question I hated in my EE degree. The thought exercise was a favourite of the profs.

We investigated Amsterdam's attempt to build a 'fair' fraud detection model (lighthousereports.com | comments) Amsterdam's efforts to develop an unbiased welfare fraud detection algorithm faced significant challenges over the years. This investigation analyses the city's initiative and the findings related to the fairness and effectiveness of the model it aimed to implement.

• Amsterdam reduced bias by one measure (False Positive Share) and bias increased by another measure (False Discovery Rate). This isn’t a failure of implementation; it’s a mathematical reality that you often can’t satisfy multiple fairness criteria simultaneously. Training on past human decisions inevitably bakes in existing biases.

Last fifty years of integer linear programming: Recent practical advances (hal.science | comments) Mixed-integer linear programming (MILP) has emerged as a crucial area in operations research over the past fifty years, driven by advances in solvers that can solve complex problems efficiently. This article reviews key developments, focusing on computational improvements and practical applications across various industries, while also addressing ongoing challenges in MILP research.

• title could use [pdf] [2024]
• I've heard Gurobi is fairly expensive. Anyone willing to share pricing details?
• Integer linear programming doesn't sound very complex.

Model Once, Represent Everywhere: UDA (Unified Data Architecture) at Netflix (netflixtechblog.com | comments) Netflix introduces the Unified Data Architecture (UDA), a knowledge-graph-based framework that transforms conceptual domain models into standardized schemas and data pipelines, aiming to enhance data consistency and accessibility across various platforms. This approach facilitates seamless data integration and management.

• How does this relate to domain-driven design? It seems to be at odds with it, because in DDD it's kind of expected that the same concept will be represented in a different way by each system? But to be honest, I didn't read the whole blog post because of the UML vibes.