Search the archive

REACT QUERY - THE BAD PARTS

A slide deck on the tradeoffs made when choosing React Query, such as it having a large bundle size.

FEATURE: DOM STATE-PRESERVING MOVE

A new DOM `moveBefore` primitive added to the latest version of Chrome (133) allows elements to be moved within a DOM tree without losing their state, preserving features like iframe content, focus, and active CSS transitions.

APPLE LAUNCHES INVITES, A NEW APP FOR CREATING CUSTOM INVITATIONS

Apple's new Invites app allows users to create, share, and RSVP to events. It also allows users to contribute to Shared Albums and create event soundtracks. The app requires an iCloud+ subscription, but anyone can RSVP to events regardless of whether they have an Apple account or device. Invites is now available for free on the App Store for all iPhones running on iOS 18 or later and can be accessed on the web.

OPENAI'S SORA FILMMAKING TOOL MEETS RESISTANCE IN HOLLYWOOD

OpenAI has been talking to the movie industry's largest studios for months, but those talks have yet to produce any deals. Studios are reluctant to get into business with an AI company as it may anger the labor unions that they work with. Screenwriters and actors have been urging studios to police tech companies and ensure that their work isn't being used illegally. OpenAI found a receptive audience among some filmmakers when it released its Sora AI video generator early last year, but conversations with studios have been more complicated. While studios recognize that AI is already changing how people work, they are reluctant to hand over their most valuable assets to large tech firms without receiving a significant payout.

WITH GENERATIVE AI, MIT CHEMISTS QUICKLY CALCULATE 3D GENOMIC STRUCTURES

MIT chemists have figured out a new method for determining 3D genome structures using generative artificial intelligence. The technique can predict thousands of structures in minutes. It can be used to easily study how the 3D organization of the genome affects individual cells' gene expression patterns and functions. It could also be used to explore how mutations in DNA sequences can cause disease. The model and all of the data are available on GitHub.

OUR BRAINS ARE SOAKING UP MICROPLASTICS MORE THAN OTHER ORGANS

Microplastics have been detected in human brains at much higher concentrations than in other organs. Much of the plastic appears to be smaller than previously thought, with some pieces only about two to three times the size of a virus. People diagnosed with dementia had up to 10 times as much plastic in their brains as others, but there is not enough evidence to definitively conclude on cause or effect. It is unknown what effect microplastics have on human health, but the tiniest pieces can cross the blood-brain barrier.

BEEJ'S GUIDE TO GIT

Git is a version control system that keeps snapshots of source code in directory trees. It allows users to make changes and then commit those changes to the source code repository once a feature is ready. Git keeps a history of all commits made, so users can look at previous commits whenever required. This page contains a guide to Git that covers the basics and how to use it. It assumes readers have basic POSIX shell usage skills and that they are working in a Unix-like environment.

HOW I USE LLMS AS A STAFF ENGINEER

This post discusses how a staff engineer from GitHub who worked directly on Copilot uses large language models. Large language models can be helpful for smart autocomplete, short tactical chains, writing use-once-and-throwaway research code, asking questions to learn about new topics, last-resort bug fixes, and proofreading. The technology still isn't great for writing PR, ADRs, or other technical communications. It also isn't great for research in large codebases.

CHAT IS A BAD UI PATTERN FOR DEVELOPMENT TOOLS

Coding forces humans to think like machines. While programming has become more friendly throughout the decades, we are still just translating human thoughts into computer instructions. AI was supposed to allow plain English to become a programming language, but it isn't suitable for building serious software. Current tools pretend that writing software is like having a conversation, but real software can't be built without being precise about requirements. Companies need to build tools for real software instead of toys.

OPENAI HAS UNDERGONE ITS FIRST EVER REBRAND, GIVING FRESH LIFE TO CHATGPT INTERACTIONS

OpenAI has unveiled a crisp new look for its brand and products. The company has, until now, presented itself with a jumble of fonts, marks, and colors that lack cohesion. The new identity is more organic and human while still presenting a sense of authority, technological sophistication, and research-focused underpinnings. It is present through the interface and presentation of all of the company's products. This article takes a detailed look at the design.

CRUISE TO SLASH WORKFORCE BY NEARLY 50% AFTER GM CUTS FUNDING TO ROBOTAXI OPERATIONS

The job cuts extend to the CEO and several other top executives.

HOW TO SCALE YOUR MODEL

Model scaling increases the number of chips used for training or inference, which usually decreases computation time, but at the cost of added communication between chips.

8 MILLION REQUESTS LATER, WE MADE THE SOLARWINDS SUPPLY CHAIN ATTACK LOOK AMATEUR

This article looks at the potential impacts that abandoned infrastructure can have.

FIGURE DROPS OPENAI IN FAVOR OF IN-HOUSE MODELS

Figure AI's CEO has promised to deliver something no one has ever seen in a humanoid robot in the next 30 days following a major breakthrough in the company's in-house AI.

WHAT'S THE BEST EMBEDDING MODEL FOR SEMANTIC CACHING?

Semantic caching enables faster, cost-effective responses for similar queries, changing how we optimize systems that rely on large language models.

SPREADSHEETS FOR EXPLORING SCENARIOS

Ambsheets introduces a feature where a single cell holds multiple values at once, allowing for the creation of spreadsheets that consider lots of possible scenarios.

HUGGING FACE REPLICATING OPENAI'S DEEP RESEARCH

Hugging Face attempted to replicate OpenAI's Deep Research, an agentic web-search framework that significantly improved performance on the GAIA benchmark, by running a 24-hour-long experiment aimed at open-sourcing an equivalent system.

GOOGLE CEO ON DEEPSEEK VS. GEMINI

Sundar Pichai has downplayed the efficiency of DeepSeek's AI models, arguing that Google's Gemini models, particularly Gemini 2.0 Flash, outperform them despite DeepSeek's disruptive impact on the AI market.

US COPYRIGHT OFFICE RULES OUT COPYRIGHT FOR AI CREATED CONTENT WITHOUT HUMAN INPUT

The US Copyright Office states that AI-generated works without human intervention cannot be copyrighted. AI tools assisting with creativity, like de-aging actors, won't limit copyright protection, but purely generative AI outputs require further analysis.

HARMONIC LOSS TRAINS INTERPRETABLE AI MODELS

Harmonic loss is an alternative to cross-entropy loss for training neural networks that offers better interpretability and faster convergence through scale invariance and finite convergence points. Experiments across algorithmic, vision, and language datasets, demonstrate that models trained with harmonic loss show superior performance in interpretability, data efficiency, and reduced grokking compared to standard models. Harmonic loss could be particularly valuable for applications with limited data or where interpretability is crucial.

CONVEX OPTIMIZATION THEORY AND LEARNING-RATE SCHEDULING FOR LARGE MODEL TRAINING

Learning-rate schedules for large models closely match theoretical bounds from non-smooth convex optimization. These authors provide a bound for constant schedules with linear cooldown, showing cooldown's practical benefits through the absence of logarithmic terms in the bound. Their findings enabled practical improvements in training Llama-type models through optimal learning-rate extension and cross-schedule transfer.

IN-CONTEXT REINFORCEMENT LEARNING

This study explores scaling In-Context Reinforcement Learning (ICRL) to broader domains using Algorithm Distillation and shows that ICRL can be a viable alternative to expert distillation for generalist decision-making systems.

Forrester report

GOT OCR 2.0 WEIGHTS (HUGGING FACE HUB)

One of the best OCR models is now available and integrated with the Hugging Face ecosystem. It works well on documents and sheet music.

OPEN-VOCABULARY DETECTION WITH LLMS

LLMDet is an open-vocabulary detector that leverages a large language model for enhanced caption generation and grounding that significantly boosts performance compared to existing detectors.

EFFICIENT CHAIN-OF-THOUGHT REASONING

Heima introduces a framework for more efficient multimodal reasoning by compressing Chain-of-Thought processes into a single hidden token.

HOW TO SCALE YOUR MODEL

Amazing post from the DeepMind team about the mental process they use to scale their model. They break it down into mathematical equations, which allows them to reason about the costs of each operation and ensure correctness.

WHO IS LIANG WENFENG? DEEPSEEK FOUNDER COMES FROM AI INVESTING

DeepSeek's R1 reasoning model uses less computing power than its U.S. counterparts and is open source. The DeepSeek app topped App Store charts over ChatGPT. Founder Liang Wenfeng previously started AI firms and his hedge fund High-Flyer manages $8 billion, backing DeepSeek. Liang sets himself apart by offering the product for free and open source.

AI AND THE FUTURE OF NATIONAL SECURITY

Google highlights the strategic importance of AI and quantum computing for national security, emphasizing the need for private-sector leadership, government procurement reforms, and public-private collaboration to strengthen cybersecurity.

GOOGLE'S 2024 RESPONSIBLE AI REPORT

Google has released its 6th annual Responsible AI Progress Report, which details governance structures, safety evaluations, and risk mitigation techniques for AI product development.

HUGGING FACE RESEARCHERS ARE TRYING TO BUILD A MORE OPEN VERSION OF DEEPSEEK'S AI 'REASONING' MODEL

Hugging Face is working to replicate DeepSeek's R1 model with an open-source initiative called Open-R1, addressing concerns over the lack of transparency in DeepSeek's release.

META AI CAN NOW USE YOUR FACEBOOK AND INSTAGRAM DATA TO PERSONALIZE ITS RESPONSES

Meta is enhancing its AI chatbot with memory capabilities that will allow it to remember user details in conversations on Facebook, Messenger, and WhatsApp in the U.S.

TRUMP'S ‘CRYPTO CZAR' FLOATS BITCOIN RESERVE AS WHITE HOUSE PUSHES SOVEREIGN WEALTH FUND

President Trump's newly appointed “crypto czar,” David Sacks, confirmed that evaluating a strategic bitcoin reserve is a top priority, aligning with the administration's broader pro-crypto agenda. This comes as Trump signed an executive order to establish a U.S. sovereign wealth fund. Analysts speculate that bitcoin could be included as a strategic asset to help bolster national reserves.

WILL TRUMP'S TRADE WAR BRING AN EARLY END TO THE BITCOIN BULL MARKET?

Bitcoin's recent drop below $92K amid Trump's trade war announcement triggered fears of an early bull market end, but strong demand and historical patterns suggest further upside. Analysts project a peak between $160K and $210K in late 2025. Key indicators, including long-term holder supply and the Pi Cycle Top, signal that Bitcoin has yet to reach its final cycle high. A potential top is expected around September–October 2025.

STORY PROTOCOL WHITEPAPER: A PEER-TO-PEER INTELLECTUAL PROPERTY NETWORK

Story Protocol is building an IP blockchain that has a programmable market for knowledge and creativity assets through a multi-core execution environment that combines EVM compatibility with specialized cores for IP management, offchain synchronization, and cross-chain communication. The platform features a Proof of Creativity protocol for IP registration with customizable usage parameters and has a native IP token for network security and transaction processing. Potential applications include a universal market for intelligence, a foundational layer for AI interactions through the Agent TCP/IP protocol, and infrastructure for autonomous agent-to-agent commerce.

BERACHAIN HONEYPAPER: PROOF OF LIQUIDITY

Berachain's novel blockchain architecture introduces Proof of Liquidity (PoL), which aims to align network security with liquidity provision. It leverages a dual-token model consisting of BERA (gas and staking) and BGT (governance and rewards). It will be an EVM-compatible chain and introduce an incentives marketplace where protocols can bid for validator emissions using whitelisted tokens, while validators can direct emissions toward reward vaults based on economic preferences. The protocol's architecture incorporates immediate execution and optimistic payload bundling to reduce block times by up to 40%. Its economic model aims to solve traditional L1 problems by adjusting validator rewards based on demand for validator security rather than preset curves.

EVOLVING COINBASE WALLET TO BRING THE WORLD ONCHAIN

Coinbase Wallet is being rebuilt to be 10x faster, easier to use, and more integrated with onchain social experiences, using Farcaster's social graph and mini-apps to enhance app discovery. The 2025 vision focuses on streamlining crypto payments, enabling builders to go viral, and creating a seamless experience for traders, creators, and communities to engage in the onchain economy.

THOUGHTS ON FAIR LAUNCHES

Current fair launch token models, which distribute 85-95% of supply at launch while leaving teams with only 5-15%, create operational challenges for long-term project development compared to traditional VC funding where teams retain 80-90% ownership. It leads to inadequate liquidity management, difficulties in team compensation, and unsustainable resource allocation for development, particularly impacting teams building complex technical solutions that require extended development cycles. While fair launches aim to democratize early access traditionally reserved for VCs, the approach often undermines project sustainability and forces builders to prioritize short-term token economics over long-term technological development.

YOU'RE PROBABLY LOOKING AT THE DEFAI OPPORTUNITY WRONG

The DeFAI market is in the midst of a serious correction, with market caps dropping over 50% from their mid-January peaks. The sector is undervalued and could grow to represent 10% of the $125 billion DeFi market. There are three value segments underpinning this belief: Information Agents focusing on UI abstraction and analysis, smart DeFi enabling automated actions, and Alpha Hunters that combine data analysis with autonomous execution. Despite market skepticism, AI-native teams are better positioned to compete in inefficient markets where returns are increasingly attributed to intelligence rather than human intervention.

CRYPTO CZAR DAVID SACKS HAS SAID NFTS AND MEMECOINS ARE COLLECTIBLES, NOT SECURITIES

White House crypto czar David Sacks argues that NFTs and memecoins should be classified as collectibles, like baseball cards or stamps, rather than securities or commodities, which would exempt them from stricter financial regulations. While his remarks don't change their legal status, they highlight the ongoing debate over how digital assets should be regulated and taxed in the U.S.

SEC LAUNCHES CRYPTO TASKFORCE

The SEC has launched a new Crypto Task Force website to provide regulatory clarity, foster innovation, and engage with the public on crypto asset laws and compliance.

BERACHAIN MAINNET SET TO LAUNCH FEBRUARY 6

Berachain's Layer 1 mainnet, set to launch on February 6, introduces a unique “proof of liquidity” consensus model that integrates staking with DeFi participation.

DEFI PROTOCOL GOLDFINCH BRINGS ONCHAIN EXPOSURE TO PRIVATE CREDIT GIANTS LIKE APOLLO WITH 'PRIME' LAUNCH

Goldfinch has launched "Prime," a new onchain investment pool that offers exposure to private credit funds like Apollo and Ares, which collectively manage over $1 trillion in assets.

ALCHEMY AND SYNDICATE JOIN FORCES TO SCALE ETHEREUM THROUGH COMMUNITY-OWNED ROLLUPS

Alchemy and Syndicate are teaming up to scale Ethereum by integrating Alchemy's rollup developer platform with Syndicate's onchain sequencing network, enabling seamless, community-owned rollups.

ENRON LAUNCHES A MEMECOIN

A new Solana memecoin, $ENRON, has launched - it embraces the infamous name of the collapsed energy giant for viral appeal.

APPLE LAUNCHES INVITES, A NEW APP FOR CREATING CUSTOM INVITATIONS

Apple has introduced Invites, a new app designed to create and send custom digital invitations with seamless Apple ecosystem integration. The app offers a variety of templates, Apple Pencil support, and iCloud syncing, making event planning more visually appealing and convenient. While initially exclusive to iOS and macOS, Apple is exploring broader compatibility to compete with established invitation platforms.

MARVEL DENIES USING AI FOR ‘FANTASTIC FOUR' POSTER

Marvel has denied accusations that its newly released Fantastic Four teaser poster was generated using AI following online speculation about its design. Critics pointed to certain stylistic elements that resembled AI-generated art, sparking debate among fans and industry professionals. Marvel insists the artwork was created by human designers, reaffirming its commitment to traditional design methods.