Search the archive

WHY DON'T MORE PEOPLE USE LINUX?

While Linux may be free, it demands more of the user than the commercial offerings from Microsoft and Apple. Most computer users aren't interested in an intellectual workout when all they want to do is browse the web or use an app. Programmers should be different - they should take the time to learn the operating system to develop a better understanding of the basics of the Internet. While it may not be easy to learn to use Linux, it's definitely worth it.

THE ART OF FINISHING

Finishing projects teaches us different lessons than starting them - the end stages of a project often involve optimization, refactoring, and really understanding the intricacies of your code.

OPENAI PLANS TO BUILD ITS OWN AI CHIPS ON TSMC'S FORTHCOMING 1.6 NM A16 PROCESS NODE

The A16 process node is still in development and is still years away from being ready for bulk orders.

THE MOST POPULAR PROGRAMMING LANGUAGES IN 2024 (AND WHAT THAT EVEN MEANS)

People looking for programming jobs should brush up on their SQL, Python, and Java.

DIFFERING VALUES IN A TEAM ARE COSTLY

It just ends up with a bunch of people being frustrated, because nobody is wrong - they just don't have the same values.

REASONS TO WRITE DESIGN DOCS

Writing design docs helps you think, leading to better designs and a better understanding of the design within the organization.

THIS WEEKEND, THE @XAI TEAM BROUGHT OUR COLOSSUS 100K H100 TRAINING CLUSTER ONLINE

It took 122 days to build Colossus - the cluster will double in size in a few months.

AI that prioritizes data privacy.

You control access to AI-powered capabilities. Your code and data are never used to train AI models.

XAI STARTS COLOSSUS TRAINING CLUSTER

xAI has started the 100,000 Colossus H100 training cluster, the largest in the world, with plans to double it in size in a few months.

WHY AI CAN'T SPELL 'STRAWBERRY'

Large language models sometimes fail at tasks like counting letters due to their tokenization methods. This highlights limitations in LLM architecture that affect their understanding of text. Nevertheless, advancements continue, such as OpenAI's Strawberry for improved reasoning and Google DeepMind's AlphaGeometry 2 for formal math.

FIRST MLPERF BENCHMARKS FOR NVIDIA BLACKWELL, AMD, GOOGLE, UNTETHER AI

Nvidia's new Blackwell chip demonstrated top per GPU performance in MLPerf's LLM Q&A benchmark, showcasing significant advancements with its 4-bit floating-point precision. However, competitors like Untether AI and AMD also showed promising results, particularly in energy efficiency. Untether AI's speedAI240 chip, for instance, excelled in the edge-closed category, highlighting diverse strengths across new AI inference hardware.

DIFFUSION MODELS PERFORM SPECTRAL AUTO REGRESSION

Diffusion models and auto regressive models are often thought to be fundamentally different forms of methods. However, they may actually be more similar than we thought where diffusion models are actually taking auto regressive steps in the frequency domain.

ENHANCED ALZHEIMER'S DIAGNOSIS

This study presents a multimodal classification model designed to improve the early diagnosis of Alzheimer's Disease.

CAUSAL INFERENCE IN COMPLEX SYSTEMS

Researchers have developed an advanced machine learning method for identifying both direct and indirect causal variables of a target within a system.

Marriott Marquis

ML DTYPES

Bfloat16 and fp8 support for native numpy arrays.

TIME SERIES FORECASTING WITH VISION MODELS

VisionTS is a new approach that leverages the similarities between natural images and time series data to enhance forecasting by reformulating time series forecasting as an image reconstruction task. It uses a visual masked autoencoder (MAE) pre-trained on ImageNet to achieve impressive zero-shot performance.

IMPROVING AUDIO GENERATION WITH SEMANTIC AWARENESS

X-Codec is a new approach to enhancing the performance of audio generation by LLMs.

CAN AI SCALING CONTINUE THROUGH 2030?

AI training is growing at an unprecedented 4x per year, outpacing mobile adoption and genome sequencing rates from past technological expansions. Research indicates scaling AI training could feasibly continue until 2030, constrained primarily by power availability and chip manufacturing capacity. Training runs up to 2e29 FLOP may become possible, marking substantial progress akin to the leap from GPT-2 to GPT-4, provided hundreds of billions are invested. Data scarcity and latency also pose challenges, but these may be surmountable through multimodal and synthetic data generation strategies and advanced network topologies.

MEET BOARDWALK ROBOTICS' ADDITION TO HUMANOID ROBOTS

Boardwalk Robotics has unveiled Alex, a humanoid upper torso robot designed for tasks in manufacturing, logistics, and maintenance. Developed independently while leveraging the pedigree of IHMC's bipedal robot expertise, Alex is legless, with its creators emphasizing manipulation over mobility to ensure safety and efficiency. Commercial partners are currently being selected for pilots, while researchers can purchase Alex immediately.

GPU UTILIZATION IS A MISLEADING METRIC

While commonly monitored, GPU utilization may not accurately reflect true GPU performance in ML workloads, as it doesn't account for whether the GPU's computational power is fully leveraged. Trainy discovered this while only achieving ~20% model FLOPS utilization (MFU) despite 100% GPU utilization during LLM training. It recommends tracking SM efficiency for a better performance indicator and achieving a 4x speedup in training time by optimizing with fused kernels and the right model parallelism level.

AI-IMPLANTED FALSE MEMORIES

A study from MIT Media Lab found that generative chatbots powered by large language models significantly increased the formation of false memories during simulated crime witness interviews, inducing over three times more immediate false memories than a control group.

VISION MODEL LEADERBOARD (HUGGING FACE HUB)

Timm is a package of vision models.

COGVIDEO 5B LAUNCHING NEXT WEEK

CogVideo 5B model will launch next week in Hugging Face Diffusers.

→

QATAR INTRODUCES REGULATORY FRAMEWORK FOR DIGITAL ASSETS

Qatar (led by the Qatar Financial Centre) has launched a regulatory framework for digital assets to increase innovation and efficiency in the country's financial sector. This framework hopes to have provisions around tokenization, custody arrangements, and the legal recognition of smart contracts and property rights in tokens, with the hope of bolstering economic development and attracting foreign investment.

WHERE DID ETH GO WRONG?

This critique of Ethereum highlights several key issues: an overemphasis on sound money principles rather than scaling, which led to slow and expensive transactions, and a lack of support for DeFi. The proliferation of L2s has fragmented liquidity, undermining Ethereum's auction-based fee model and its sound money thesis, while ongoing selling pressure from the Ethereum Foundation and competition from user-friendly alternatives like Solana exacerbate the challenges.

QUICK TLDR OF ETHER VISTA

A new DEX, similar to Uniswap v2 but with a unique fee structure, has quickly gained attention, reaching $27 million. Users must create their own tokens, while the DEX provides a "Launch" function, distributing LP tokens in exchange for new tokens and ETH, with fixed buy/sell fees that reward LP providers and the token creator.

BRIDGE: A NEW STABLECOIN PAYMENTS PLATFORM

Bridge uses stablecoins to address global financial challenges caused by local currency limitations. It aims to simplify cross-border transactions, provide dollar access in unstable economies, and reduce costs associated with traditional money movement. Bridge has partnered with organizations to facilitate payouts, cross-border payments, and FX operations by offering stablecoin orchestration and issuance.

WHY SEQUENCING WILL ALWAYS ACCRUE WAY MORE VALUE THAN DA

The Ethereum rollup-centric roadmap is flawed because it undervalues the sequencing layer, which captures most of the value through price discrimination, while the data availability (DA) layer, like Ethereum's, earns little since it can't price discriminate. This misjudgment has led to a loss of value accrual to the base layer, threatening Ethereum's role as network equity rather than as a purely monetary asset.

A GUIDE TO ZKTLS

In zkTLS, a system for creating Web Proofs that can bring Web2 to Web3, zero-knowledge proofs are used not for privacy but to ensure integrity and security. There are three main approaches to zkTLS - TEE-based, MPC-based, and Proxy-based - each using different cryptographic primitives to protect the private key while maintaining claim integrity. The primary purpose of these systems is to prove that data came from a specific website without revealing the decryption key.

GLUE AND COPROCESSOR ARCHITECTURES

This article explains how modern computing often separates tasks into two parts: simple logic that directs the process ("glue") and intense, specialized work ("coprocessors"). For example, Ethereum uses general code for business logic and specific code for heavy tasks like cryptography. This trend is growing because it balances efficiency with flexibility, making it easier for developers to build complex systems while maintaining performance and security. The approach also opens doors for new, smaller developers to contribute effectively.

THE ‘GOLDEN YEAR OF AIRDROPS' IS OVER

Airdrops have become institutionalized and overfarmed, which, alongside disdain towards VC-backed projects, has made airdrop farming and token releases less valuable. There will still be under-the-radar projects that adequately reward users, but it is worth seeking those out and staying at the cutting edge of protocols.

THE FALSE DEMISE OF THE EVM

Ethereum and the EVM still dominate the developer ecosystem, with most new developers joining the Ethereum network, its L2 solutions, and EVM-compatible chains like Polygon, BNB, and Avalanche.

SOLANA VALIDATOR REVENUE BREAKDOWN

Solana validators receive 78.2% of revenue from issuance, 15.3% from MEV, and 6.5% from fees.

IF THIS THEN THAT FOR THE INTERNET

This post discusses the potential of a futuristic "If This, Then That" (IFTTT) mechanism for creating automated actions in the blockchain space that allows users to set up smart contracts that execute based on verified real-world events.

ARE CEX LISTINGS REALLY A PRICE CATALYST

While crypto listings on major exchanges like Binance are often viewed as price catalysts, the actual impact on token prices post-listing is quite mixed.

NIKE HAS A SECRET DESIGN ARCHIVE. THIS NEW MUSEUM SHOW TAKES YOU INSIDE OF IT

The Vitra Design Museum's "Nike: Form Follows Motion" exhibition showcases over 200,000 rare Nike artifacts from the brand's secret archive, the Department of Nike Archives (DNA). The exhibit explores Nike's history, innovative sports designs, and future direction, showcasing the company's creative approach and groundbreaking research initiatives.

MOVE OVER MIDJOURNEY — FREEPIK MYSTIC IS THE MOST REALISTIC AI IMAGE GENERATOR I'VE TRIED

Freepik and Magnific AI have released Freepik Mystic, a new AI image generator that accurately creates hyperrealistic visuals following complex prompts. It's built on top of a series of fine-tunes of Stable Diffusion, Flux, and Magnific-built models. Offering high-resolution (2k) outputs, it is currently exclusive to Freepik premium users, but a wider rollout is anticipated.

NIKE'S OLYMPIC GAMES TYPEFACE HAS A LOT OF MUSCLE

WHY AI'S LACK OF DIRECTION IS PERFECT FOR CX'S LOSS OF DIRECTION

AI's lack of inherent direction aligns perfectly with Customer Experience's (CX) current struggles because it allows for flexibility and experimentation. AI's focus on solving simple, repetitive tasks contrasts with CX's outdated problem-solving models, creating opportunities for unexpected, innovative approaches that redefine CX by combining insights from diverse fields.

THINGS TO DO BEFORE ASKING “IS THIS ACCESSIBLE?”

Understanding context is crucial when evaluating accessibility. Key questions can be adapted for different products or services, forming the basis of a procurement checklist. A balanced approach that combines automated tools with manual reviews while also focusing on the intended audience and fostering learning in accessibility is essential.

FOCUS YOUR CV ON YOUR SKILLSET AND THE THINGS YOU ENJOY DOING

To avoid being seen as a generalist, focus on skills, enjoyable tasks, and achievements rather than job titles. Tailor CVs to each role, use LinkedIn to understand job requirements, seek mentors, join networks, and showcase leadership and project examples for senior roles.

APP MOTION

AppMotion is a mobile motion design gallery that showcases the best, hand-picked examples of app motion design. The gallery allows you to explore various design elements, such as checkboxes, pinch, spin, morph, and more.

RESEARCH VISUALLY ON INFINITE CANVAS

Noteey is a visual learning tool that offers an infinite canvas for organizing and connecting ideas, annotating PDFs, and creating mind maps.

INSTANTLY TURN YOUR FIGJAM INTO A WEBSITE

Figgy allows users to quickly build and maintain websites from Figjam files with a drag-and-drop interface. It offers instant updates and doesn't require any coding or technical experience.