Search the archive

RITUALS FOR GREAT DECISION MAKING

A strong decision-making ritual includes four key steps: ask the right questions, follow the right process, choose the right forum, and engage the right stakeholders.

THE CONTEXT MATRIX

In product management, creating context involves gathering data on market trends, capabilities, and strategic goals, while making sense of it relies on the team's skills to interpret and act on this data.

ALL THE DATA CAN BE YOURS

This post goes over various methods for discovering and utilizing undocumented APIs to reverse-engineer web apps. These techniques range from searching GitHub for public code and analyzing fetch requests and robots.txt files to leveraging Google's cached errors and guessing API endpoints based on patterns.

OPTIMIZING A WEBGPU MATMUL KERNEL FOR 1TFLOP+ PERFORMANCE

This article discusses the optimization of a WebGPU matrix multiplication kernel that achieved over 1 TFLOPS performance. It explores each stage of the kernel optimization, starting with a naive implementation and progressing through increasing thread count, 2D workgroups, kernel tiling, and loop unrolling. Through these optimizations, the kernel achieved a performance improvement approaching the theoretical peak of the Apple M2 Pro chip.

HOW TO SHIP PROJECTS AT TECH COMPANIES

Shipping projects in big tech companies requires a unique skillset that goes beyond just writing code. The most important aspect of shipping is gaining the trust and approval of company leadership, which means understanding their expectations, communicating well, and proactively addressing potential problems.

SECURITY IS A USELESS CONTROLS PROBLEM

Many security controls are useless because they lack clear, understandable explanations. These controls often arise from compliance requirements or security "best practices" that don't consider their actual value.

ARE AI ASSISTANTS MAKING US WORSE PROGRAMMERS?

While AI can help with repetitive tasks and code generation, over-reliance can lead to a deterioration of fundamental skills. AI assistants are most beneficial for handling implementation details and tedious aspects of coding so that devs can focus on problem-solving, design, and strategic thinking.

STDILB

stdlib is a standard library for JavaScript and Node.js that focuses on numerical and scientific computation with a fully decomposable architecture.

SKELETON

Skeleton is an adaptive design system for Tailwind that provides opinionated solutions for themes, colors, typography, and more. It offers both Tailwind-based and framework-specific components.

WEBLLM CHAT

WebLLM Chat is an open-source, private AI chat interface that runs large language models natively in browsers using WebGPU, ensuring privacy and offline accessibility.

INSIDE THE $20 MILLION BUSINESS OF GUTTING FAILED BAY AREA TECH COMPANIES

Better Source is a $20 million business that specializes in liquidating office furniture from failed Bay Area tech companies. The company has witnessed the rise and fall of numerous tech giants. Its business has boomed in recent years because of the shift to remote work and the changing demands for office spaces.

THE DEATH AND LIFE OF PREDICTION MARKETS AT GOOGLE

Google has experimented with two internal prediction markets, Prophit and Gleangen. Prophit, launched in 2005, was initially popular but ultimately failed to launch publicly due to legal challenges and a shifting regulatory landscape. Gleangen, launched in 2020, has been much more successful internally, with a larger user base and more consistent use, but faces ongoing challenges in proving its value to Google management.

REVERSE ENGINEERING: FINDING EXPLOITS IN VIDEO GAMES

You can reverse engineer game data by extracting and decrypting files and scripts to find exploits in video games. You can also analyze network traffic using tools like AssetsTools.NET, Fiddler, and Wireshark. By analyzing the game's authentication flow, packet structures, and opcodes, you can identify vulnerabilities and test for potential exploits.

A NEW CHAPTER, AND THOUGHTS ON A PIVOTAL YEAR FOR C++

Herb Sutter, a prominent C++ expert, is leaving his position at Microsoft Visual C++ compiler team to join Citadel Securities, where he will be working on C++ training initiatives and advising on technical strategy.

NO MORE REDUNDANT ENGINEERING DEBATES: CREATING ALIGNMENT AND CLARITY WITH ADRS

Architectural Decision Records (ADRs) are a structured way to document key decisions, align teams, and prevent repetitive debates.

MISGUIDED APPLE INTELLIGENCE ADS

Apple's new ads for Apple Intelligence portray the features as tools for laziness, rather than highlighting their potential to help people be more productive and thoughtful, missing an opportunity to showcase the service's positive aspects.

WHY DO DEVELOPERS LOVE CLEAN CODE BUT HATE WRITING DOCUMENTATION?

Despite loving clean code, developers hate writing documentation - AI-powered tools can automate documentation creation, which is a win-win.

Invest in RYSE alongside the Dragons.

OPENAI REPORTEDLY DEVELOPING NEW STRATEGIES TO DEAL WITH AI IMPROVEMENT SLOWDOWN

OpenAI's new model's performance exceeds its existing models, but the improvement is less than the jump from GPT-3 to GPT-4. The rate of improvement seems to be slowing down. OpenAI created a foundations team to figure out how the company can continue to improve its models. One of the proposed strategies is to train models on synthetic data produced by AI models.

CHATGPT MONTHLY USAGE MAY NOW RIVAL GOOGLE CHROME

SPACEX PITCHES NASA ON 'MARSLINK,' A VERSION OF STARLINK FOR THE RED PLANET

SpaceX is working on a concept version of Starlink for Mars. It has ideas about developing next-generation relay services capable of beaming 4 Mbps or more in data across 1.5 astronomical units. The Marslink concept involves placing multiple space satellites in Mars orbit to provide full visibility and interoperability for ground and orbital assets. Other companies have also submitted proposals - slides from the recent meeting of the Mars Exploration Program Analysis Group showing each company's proposal are available in the article.

HPV VACCINATION: HOW THE WORLD CAN ELIMINATE CERVICAL CANCER

Cervical cancer is caused by a virus (HPV), making it preventable. The virus is also responsible for a large share of other cancers. HPV integrates itself into cells' DNA and damages key proteins that protect the cell from uncontrolled growth, eventually leading to cancer. HPV vaccines are highly effective in preventing infections - and cervical cancer. Countries like the UK and Australia are showing that the elimination of cervical cancer is possible, but much of the world is missing out.

constantly. Deploy continuously. Start here

EVERYTHING I'VE LEARNED SO FAR ABOUT RUNNING LOCAL LLMS

It is now possible to run a large language model (LLM) smarter than the original ChatGPT on a Raspberry PI. Running models locally allows for private, unlimited, and registration-free access. This article talks about how to run local text-based LLMs on your own hardware. It covers the software needed, the different models available, user interfaces, and more. While LLMs are fun, there are still many tasks they can't reliably do.

IRONCALC

IronCalc is a modern spreadsheet engine and set of tools for working with spreadsheets in diverse settings. It can be used with a variety of programming languages including Python, JavaScript, Node.js, and possibly R, Julia, and Go. IronCalc is still a work in progress but an early browser preview is available.

MOVING PAST SKEUOMORPHIC AI

AI is being limited by our mental models for what to use the technology for. The current available technology can already be transformative given the right application mindset. Moving to the next level will require giving models more autonomy and self-learning capabilities. We should be building applications around things that AI is uniquely suited to do. Thinking this way will result in the creation of new services and products.

TOUCH SCREENS ARE OVER. EVEN APPLE IS BRINGING BACK BUTTONS

Screen interfaces require people to look at them, making the 'touch screen' a misnomer. Buttons, knobs, sliders, and other physical controls are making a comeback in vehicles, appliances, and personal electronics. Engineers and designers are tapping into people's sense of touch and spatial awareness to make using gadgets easier, more fun, and in some cases, safer. Physical interfaces allow people to use devices without looking and makes interactions with machines more intuitive.

to handle your customers' messiest spreadsheets. See OneSchema

SYSTEM TESTS HAVE FAILED

System tests in theory offer great confidence that machines are working as they should, but in practice, they are slow, brittle, and full of false negatives.

THIS SCIENTIST TREATED HER OWN CANCER WITH VIRUSES SHE GREW IN THE LAB

Beata Halassy, a virologist at the University of Zagreb, treated her own breast cancer by injecting the tumor with lab-grown viruses, sparking discussions around the ethics of self-experimentation.

INTERNAL SOFTWARE BUILDERS

AI could give the No Code/Low Code sector a new lease on life.

CHONKIE

Chonkie is a lightweight RAG chunking library that can be installed, imported, and run without users having to worry about dependencies, bloat, speed, or other factors.

ALGORITHMS WE DEVELOP SOFTWARE BY

A look at different programming techniques and tricks and how they help improve code quality.

AMD MAKES BIGGEST LEAP IN RECENT HISTORY

AMD gained a substantial 5.7 percentage points of share of the desktop x86 CPU market and raked in a strong increase in revenue share in the third quarter.

Read the whitepaper

MODAL BUYS TIDBYT

The first acquisition by the elastic scaling GPU company. Modal bought Tidbyt, an NYC-based hardware company, for the in-house talent of the team around infrastructure and containerization.

OPENAI FACES AI IMPROVEMENT SLOWDOWN

OpenAI's next major model, code-named "Orion," reportedly shows less improvement over previous models, suggesting a slowdown in AI advancement. To tackle this, OpenAI has formed a foundations team focused on enhancing models with alternative methods, such as synthetic data training and post-training adjustments, as the availability of new data has decreased.

WORLD'S LARGEST OPEN-SOURCE MODEL

Near Protocol has announced plans to create a massive 1.4 trillion parameter open-source AI model. It aims to surpass existing models like Meta's Llama.

FRONTIER MATH BENCHMARK

Epoch AI has released a new challenging mathematics benchmark. Most frontier models are unable to solve more than 2% of the problems.

BITNET A4.8: 4-BIT ACTIVATIONS FOR 1-BIT LLMS

One of the key challenges with 1.58bit LLMs was the lack of hardware acceleration support. This work proposes 4.8bit activations to take advantage of the INT4/FP4 kernels in new hardware. It comes at no runtime cost.

SUPERCHARGING CLIP WITH LLMS

LLM2CLIP combines CLIP's visual and textual alignment with the advanced language understanding of LLMs.

TORCH COMPATIBLE MUON OPTIMIZER

Muon was the optimizer used for the GPT-2 training record. It is a momentum adapted SGD style method. This repository contains an implementation that can be dropped in for AdamW.

MOCHI VIDEO MODEL WITH OPTIMIZED INFERENCE

Mochi is the best open source text-to-video model. On launch, it took 8 H100s to run. Now, thanks to the community, it can be run on a single 48GB L40 with no loss in quality.

TRAINABLE PYTORCH REPRODUCTION OF ALPHAFOLD3

Protenix is a working and trainable reproduction of AlphaFold3, a protein folding project from DeepMind. It was written by Bytedance's 'AI for Science' team.

LLAMAPREVIEW

LlamaPReview is an AI assistant for GitHub that offers one-click installation and auto-reviews of pull requests with context-aware analysis. It supports multiple programming languages and integrates with GitHub Actions, providing insightful feedback directly on PRs. Currently available for free, it enhances code quality by identifying issues and suggesting optimizations.

SMOLLM2

Hugging Face's SmolLM2 is a compact language model family with sizes ranging from 135M to 1.7B parameters trained on 11 trillion tokens. These models efficiently run on-device and support various tasks, with weights available under the Apache 2 license. Quantized models like the 1.7GB and 138MB versions provide diverse flexibility for different computational needs.