Search the archive

PINTORA

Pintora is an extensible JavaScript text-to-diagrams library that works in both browsers and Node.js.

INSIDE THE CULT OF THE HASKELL PROGRAMMER

Haskell is a functional programming language with elegant syntax but a steep learning curve. Although not widely adopted for practical programming, Haskell has influenced modern programming languages, with many incorporating its concepts to improve bug prevention, code correctness, and parallel processing capabilities.

BINARY SEARCH VS. PROLLY SEARCH

Binary Search is a common search algorithm with logarithmic time complexity, often used on sorted data. Prolly Search is a faster alternative that leverages the uniform distribution of data to make more intelligent guesses than simply dividing the set in half with each iteration. However, Prolly Search is not always applicable because it relies on the assumption of uniform data distribution, and it can regress to linear time complexity in certain cases.

GPUS GO BRRR

These authors, from Stanford University, focused on optimizing AI's compute usage and developed ThunderKittens, a DSL embedded in CUDA, to write efficient kernels. ThunderKittens simplifies the process of utilizing hardware features like the Tensor Memory Accelerator (TMA) and warp group matrix multiply accumulate (WGMMA) instructions, leading to significant performance improvements in Flash Attention and Based linear attention kernels.

AMAZON S3 WILL NO LONGER CHARGE FOR SEVERAL HTTP ERROR CODES

Amazon S3 is changing its billing policy so that bucket owners will no longer be charged for unauthorized requests that return a 403 (Access Denied) error and other codes.

APPLE AND GOOGLE DELIVER SUPPORT FOR UNWANTED TRACKING ALERTS IN IOS AND ANDROID

Apple and Google have partnered to create a specification for Bluetooth tracking devices that will send alerts to both iOS and Android users if an unknown tracker is detected moving with them.

COLORANG! A (SLIGHTLY) NEW WAY TO VISUALIZE RGBA COLOR

A color picker that linearly blends RGBA color to a gray value that you can specify on a slider.

can do it with _INCOGNI_ [4]. It's a personal data removal tool

that deletes your information from over 180 data brokers [4] and people search sites, where anyone can find your or your relatives' information.

WHAT WILL OPENAI ANNOUNCE MONDAY? QUITE POSSIBLY AN AI VOICE ASSISTANT

OpenAI's big announcement today may be a voice assistant. The company is reportedly working on a technology that rolls audio, text, and image recognition into one product. OpenAI will be holding a livestream at 10 am PT to announce the product. Google is expected to announce updates to Gemini at its developer conference tomorrow.

APPLE, NETFLIX, AND AMAZON WANT TO CHANGE HOW THEY PAY HOLLYWOOD STARS

Apple will soon begin basing pay on how a series or movie performs. Talent would receive bonuses on a points system, with the size of the bonuses being based on the number of people who signed up for Apple TV+ to watch, how much time they spent viewing, and the cost of the program relative to the size of its audience. Netflix and Amazon have also spent months developing plans for performance-based compensation, but neither have gone public yet with their models. Streaming services are becoming increasingly more transparent - the practice will make it easier to tie pay to performance.

COMMERCIAL SPACE STATIONS APPROACH LAUNCH PHASE

A number of private-sector companies are aiming to harbor commercial activities in low Earth orbit (LEO) as the International Space Station approaches its end of service. This article discusses the various projects, with a focus on Blue Origin's Orbital Reef program, which recently aced testing milestones for its critical life support system with assistance from NASA. Orbital Reef plans to build a space habitat with NASA as a main client starting out, with a wider goal of fostering a sustainable commercial presence in LEO. The initiative is backed by NASA's Commercial LEO Destinations program, which is providing funding to encourage the private sector to build space habitats.

'TUNGSTEN WALL' LEADS TO NUCLEAR FUSION BREAKTHROUGH

Encasing fusion plasma inside tungsten allowed a tokamak in France to set a new record by sustaining plasma at about 50 million degrees Celsius for six minutes. Tungsten has an extraordinarily high melting point. Using a tungsten-wall environment is for more challenging than using carbon, but the material allows physicists to sustain hot plasmas for longer and at higher energies and densities. While research into nuclear fusion has been slow, the field has made significant strides over the last few years.

POPOVER API

The Popover API allows developers to easily display popover content on top of other page content. It has a standard, consistent, and flexible mechanism that creates popover content that can be controlled declaratively using HTML attributes or via JavaScript. This documentation covers how to use the Popover API with examples.

AVOIDING THE SOFT DELETE ANTI-PATTERN

Programmers hate deleting things, leading people to advocate for soft deletion, where instead of really deleting a record, you add a field that marks the record as deleted. This is generally a bad idea and there are a number of better ways to ensure access to old data. This article discusses why soft deletion is a problem and discusses several alternatives that may better meet the needs of developers.

AN ODE TO EARLY EMPLOYEES

Being a founder used to be a fancy way of saying 'unemployed'. The environment was very different back in the 90s, so those who had the courage to start something were celebrated once they succeeded. Founders today are still treated as if in a class of their own, but things are very different. Virtually any founder can raise millions of dollars before there's a product or revenue, and creating a prototype has never been easier. This article discusses how things have changed and how that has affected both founders and their employees. It also provides suggestions on how founders can start building trusting relationships and celebrate those who help enable their companies in their earliest days.

ANALYST ON STARLINK'S RAPID RISE: “NOTHING SHORT OF MIND-BLOWING”

SpaceX's Starlink satellite Internet business is now profitable. The company's first batch of 60 satellites was launched just five years ago. It will have an estimated free cash flow of about $600 million this year. The business is now self-sustaining and there's still room for significant growth.

your app. Save months of dev time with OneSchema

LESSONS LEARNED REINVENTING THE PYTHON NOTEBOOK

IS DARK MATTER'S MAIN RIVAL THEORY DEAD? THERE'S BAD NEWS FROM THE CASSINI SPACECRAFT AND OTHER RECENT TESTS

Recent data suggests that Mond, as presently formulated, is no longer a viable alternative to dark matter.

SUPERFILE

superfile is a modern terminal file manager.

WAYMO SAYS ITS ROBOTAXIS ARE NOW MAKING 50,000 PAID TRIPS EVERY WEEK

Waymo has had over one million rider-only trips across four cities.

HOW TO PRICE A DATA ASSET

The value of data comes from the value of what can be done with it - every discussion of price has to begin with understanding how the data will be used and by whom.

GPUS GO BRRR

This post looks at what a team at Stanford learned by making GPUs run more efficiently and how what they learned has changed the way they think about AI compute.

output

What are you waiting for? GenAI is moving quickly...and so should you!

OPENAI CONFIRMS MAY 13 EVENT FOR ‘SOME CHATGPT AND GPT-4 UPDATES'

OpenAI has announced a live stream event scheduled for May 13 to present updates related to ChatGPT and GPT-4, possibly including the launch of an AI-powered search engine.

COHERE LAUNCHES COMMAND R FINE-TUNING

Command R fine-tuning offers industry-leading performance at a fraction of the cost. Command R with fine-tuning consistently outperforms larger models across key performance metrics that matter most for businesses. Command R fine-tuning is immediately available for businesses and developers on Cohere's platform and Amazon Sagemaker.

ALTERA RAISES $9M TO DEVELOP AI FOR DIGITAL HUMANS

Altera is building digital human beings, starting with agents who can play any game with you, just like a friend. The company raised $9 million in an oversubscribed seed round, co-led by First Spark Ventures (Eric Schmidt's deep-tech fund) and Patron (the seed-stage fund co-founded by Riot Games alums).

BLIP 3

Salesforce has trained and released the 3rd non-commercial version of the popular BLIP models, vision and language models mainly used for image understanding and captioning.

FISHING FOR MAGIKARP: AUTOMATICALLY DETECTING UNDER-TRAINED TOKENS IN LARGE LANGUAGE MODELS

Language models rely on separately trained tokenizers. These may produce tokens that are never seen during language model training. There are many in even the most powerful modern language models. This paper explores this phenomenon and presents their findings for identifying and dealing with these tokens.

THUNDERKITTENS CUDA DSL

Hazy research has introduced a new DSL for writing CUDA kernels. It has written flash attention, which is 30% faster and fits in only 100 lines of code.

Delphi Labs. Join the tech talk here

ANYTHINGLLM

All in one LLM and RAG serving framework.

MIRAGE TENSOR OPTIMIZER

The Mirage project is a multilevel tensor optimization engine that can produce extremely high performance kernels for linear algebra modules.

UNLEARNING IN RECOMMENDER SYSTEMS

E2URec is a new approach that enables large language model-based recommendation systems to forget user data efficiently and effectively, ensuring privacy without sacrificing performance.

UNIFIED TEXT TO ANYTHING MODEL

Lumina is a project that aims to be a unified text-to-X generation model. It's trained on images, video, audio, and text interleaved, which improves downstream performance.

SUPERHUMAN?

The quest for AGI is a focal point for top AI labs, with predictions for its achievement ranging from 2027 to 2047. Despite AI not yet reaching AGI, certain systems demonstrate superhuman skills in specific tasks, suggesting that AI's current best use is as a co-intelligence, augmenting human efforts rather than replacing them.

ASTERISK/ZVI ON CALIFORNIA'S AI BILL

California's SB1047 bill proposes regulations for AI models with computational capacities over 10^26 FLOPs. It focuses on ensuring these models are used safely by requiring secure environments, quick deactivation capabilities, and rigorous misuse potential testing. The bill targets only high-risk scenarios, aiming to balance innovation with safeguards against misuse in response to concerns about AI's potential impact on society.

tech expertise required! Join free on May 15

AUDIO-DIFFUSION PYTORCH

This repository is from one of ElevenLabs' research scientists and contains a potential starting point to replicate the amazing results.

AMAZON LAUNCHES BEDROCK STUDIO

Amazon's Bedrock Studio is a web-based platform within Amazon's Bedrock generative AI environment designed to help developers experiment with, fine-tune, and deploy AI models.

CLEANLAB TLM PLAYGROUND

Cleanlab created a playground for its Trustworthy Language Model.

BINANCE GETS APPROVAL FOR INDIA RETURN

Binance has secured approval to re-enter India from the country's Financial Intelligence Unit. Both Binance and KuCoin were banned from operating in India earlier this year due to non-compliance issues. The two exchanges are two of the first off-shore exchanges to re-enter the Indian market.

KRAKEN FIRED BACK AT THE SEC WITH ITS FINAL BRIEF IN THE MOTION TO DISMISS THE SEC'S COMPLAINT

Kraken has responded to the SEC's complaint with a final brief arguing for dismissal, criticizing the SEC for failing to define a specific investment contract trading on the platform and instead citing vague investment-like "expectations" and "concepts." The exchange insists on the necessity of a concrete contract for an investment contract, challenging the SEC's broader interpretations and advocating for clearer regulatory frameworks.

FULLDAS: TOWARDS MASSIVE SCALABILITY WITH 32MB BLOCKS AND BEYOND

Ethereum's original danksharding plan required 32MB blocks, which is not possible with the current network stack. Data availability sampling can be broken up into its core components, creating an efficient mechanism for making data available and having nodes use sampling to efficiently check this data. LiteDAS is proposed as a low-latency, bandwidth-efficient sampling protocol. Improvements are still needed for connecting peers for data dispersal and sampling.

EXPLORING CONSENSUS WITH PARALLEL PROPOSALS: THE DIFFERENCE BETWEEN PBFT AND BBCA-CHAIN

PBFT and BBCA-Chain are Byzantine Fault Tolerant consensus protocols that share similarities in their three-phase flow but differ in how they handle view changes and achieve parallelism. BBCA-Chain places consensus inside a DAG, allowing for parallel block transmission and reduced communication complexity compared to PBFT, which relies on a single leader to propose blocks sequentially. BBCA-Chain is most similar to Bullshark, which is used in Aleo.