We’ve talked about precision before, that’s how much space it takes to store a number. And how many unique values that number can represent. When you reduce the precision, AI models take less space on the computer, but weird things start to happen because you represent fewer numbers for the parameters. This paper studies what happens when you quantize before training at scale. They start to poke at the magical 6B parameter performance barrier.
Creating realistic talking digital people that sync with any speech audio is an important challenge in the digital world. GeneFace++ is a new method that improves lip-syncing, video quality, and system efficiency, making it the first of its kind to achieve real-time talking face generation.
Large AI models can be expensive and bad for the environment, so researchers have found a more efficient way to create smaller models using a method called knowledge distillation. This approach is often better than spending money on extra data and helps create compact models that still perform well in various language tasks.
This research introduces 'Occ-BEV', a new pre-training method for autonomous driving that focuses on making better use of multiple camera views, rather than just one, thus capturing a more accurate understanding of the surrounding environment. The technique uses a unique 3D decoder to predict the 3D geometry of the environment.
The new GPT4Tools system can manage multiple visual foundation models by analyzing language content, allowing it to automatically decide, control, and use different models based on the user's needs during a conversation. Uniquely, it enables users to teach their own language model how to use tools more effectively using a simple self-instruction and LoRA approach, making conversations about image-related tasks much smoother and more efficient.
Let's discover how to use these damn probabilistic models in the best ways possible, without sacrificing the necessary software design building blocks. Join the MLOps Community and >50 speakers
The recent MosaicML MPT models are still gaining popularity. One of the big contributions was the “Storywriter” model which could use up to 65k tokens in context. Inference with this model was painfully slow. By adding a Key/Value cache we can speed up generation by almost 10x.
Hamilton Nolan argues that CEOs are becoming increasingly unnecessary as AI becomes more sophisticated. He proposes that we should automate the CEO position and distribute the power and wealth that CEOs currently control to other stakeholders, such as workers and shareholders.
TikTok accounts are posting horrifying artificial intelligence-generated clips of murder victims — mostly children — describing their own ghastly demise.
Amazon is rolling out artificial intelligence across a dozen of its largest warehouses to screen items for damage before orders are shipped to customers.
The community governing MakerDAO has approved the purchase of up to $1.28 billion in US government bonds via BlockTower Capital. The vote supports the opening of a new real-world asset vault funded by Maker's overcollateralized DAI stablecoin. This move is part of Maker's strategy to diversify the reserve assets backing DAI and generate revenue through yield strategies. It also highlights the growing demand for traditional financial instruments among crypto-native entities like DAOs.
Binance will delist 12 privacy coins, including Zcash and Monero, in France, Italy, Poland, and Spain starting June 26. Privacy coins enable anonymous transactions, making tracking sender, recipient, and transaction amounts difficult, which has led to the EU’s concerns over potential money laundering. Although Binance aims to support numerous projects, it must comply with local laws and regulations. Affected users have been informed that they will be unable to purchase or trade privacy tokens on the platform after June. Withdrawals will remain open.
Nike and EA Sports are teaming up to integrate digital collectibles of Nike merchandise into future EA games via .Swoosh, Nike's web3 platform that allows users to buy membership cards and digital sneakers. The aim of the collaboration is to enhance in-game experiences with a focus on customization, enabling players to acquire virtual Nike items for their in-game characters. While the specifics of the partnership are yet to be disclosed, Nike’s platform has demonstrated its potential with a fairly limited offering, generating close to $200 million in revenue since its launch in December 2021.
LayerZero has announced the launch of a production-grade zk light client developed by the PolyhedraZK team. This zkLightClient is based on notable work in the field of zero-knowledge proofs and is built exclusively on LayerZero. It features custom circuits that achieve significant compression and can batch multiple transaction verifications into a single zk proof. LayerZero's unique modular configuration allows every application to construct its desired security stack, thus offering a flexible security-cost balance tailored to individual applications.
This blog from Helius dives into Solana's Token2022, an upgrade to its current token program. It introduces extensions for enhanced functionality while maintaining functionalities like minting, transferring, and burning tokens. New features include transfer fees, transfer hooks, closing mint, interest-bearing tokens, non-transferable tokens, confidential transfers, memos on incoming transfers, immutable ownership, default account state, and permanent delegate. These additions meet the growing demand for more complex and adaptable token features, providing developers more flexibility. Despite similar features on Ethereum and Binance Smart Chain, Solana's speed, low transaction costs, and Token2022's new functionalities could provide a competitive edge in the L1 race.
The article discusses the concept of "intents" in Ethereum, which refer to the desired outcomes of transactions, an innovation which could enable an improved user experience by simplifying transactions. Intents are sets of declarative constraints that allow users to outsource transaction creation to third parties without losing control. Although intents can provide enhancements to the Ethereum ecosystem, they also pose challenges, particularly with off-chain infrastructure. These challenges include the potential for centralization and the proliferation of middlemen, order flow control, and opacity.
Coinbase Wallet has introduced zero gas fees for USDC transactions on the Polygon network, enhancing convenience and efficiency. This advancement has the potential to further simplify cross-border transfers and makes USDC operations easier. Enabled by EIP 4337, or "Account Abstraction", developers can subsidize gas fees for certain account transfers. Some Coinbase users hope this zero gas fee offering could extend to other networks like Solana, NEAR, or Avalanche. In other USDC-related news, Circle announced that USDC will be natively available on Arbitrum on June 8.
Ledger Live and Ledger hardware wallets are now compatible with ENS names. This can help simplify the user experience and enhance transaction security. The integration utilizes Ledger's new Trusted Name Service, which ensures secure mapping between ENS and Ethereum hexadecimal addresses. Since Ethereum’s hexadecimal addresses are long and complicated, ENS enables users to replace these addresses with simpler, human-readable aliases, reducing transaction errors. With this update, the hardware wallet’s display can now show these ENS addresses, which further facilitates easier transaction verification. Currently, only ENS is supported, but according to Ledger, compatibility with more networks and namespace providers will be added in the future.
‘The Goose’, one of the most recognizable NFT pieces that exists, is part of Dmitri Cherniak’s Ringers collection. It was purchased by 3AC for $5.9m and is now being auctioned off by Sotheby’s as part of the defunct firm’s liquidation process with an estimated value of $2m - $3m, which a Sotheby’s VP said is “conservative”.
Redeemable NFTs offer additional utility beyond being digital art or membership into a club, serving as digital assets that can be exchanged for physical items, event access, other NFTs, or various experiences, bridging the gap between the virtual and physical world. More and more products enabling these types of NFTs have been cropping up with good initial traction.
A Redditor posted an analysis of why he thinks Lido’s market share is a threat to the entire ETH ecosystem. LDO holders have voted against self-limiting inbound stake flow. This means that the only real option to combat this large stake share is for other protocols to crop up and take share.
Giddy has partnered with Bitrefill to allow users to use DeFi yield for real-world purchases, offering auto-compounding and a variety of payment options across 5,600+ retailers.
It takes, on average, 8 months for new hires to become productive, however the best companies can reduce that duration to 3 months, this saving $50,000 for a $120,000 employee.
Jason Lemkin breaks down his top 20 learnings on public SaaS companies. As a group, they are all growing faster than last year, at an average of $400m in ARR.
This article is an in-depth exploration of rate limiting. Rate limiting restricts the number of times users or services can access a resource such as an API or network. The piece delves into the uses, benefits, and applications of rate limiting. It also examines its key design considerations and popular algorithms, like Fixed Window Counter and Sliding Window Log. The article explains the core concepts involved in rate limiting implementations and describes the types of rate limiting responses.
This post presents a simple way to generate maps entirely within a Postgres database without any external infrastructure. It uses SVG to store and display image data.
Most people think assembly is only to be used to write toy programs for learning purposes or to write a highly optimized version of a specific function inside a codebase written in a high-level language. This tutorial goes through how to write a GUI from scratch in Assembly. It covers the basics of Assembly, sockets, server connections, graphical contexts, and more.
Zig is a general-purpose programming language and toolchain for maintaining robust, optimal and reusable software. This article goes into the reasons Zig is challenging to learn: it’s new, it forces you to make choices, it’s pedantic, and it has compile time execution. However, the author believes that regardless of the difficulty in the beginning, Zig is definitely worth using and learning.
As a developer, it’s hard to estimate how long a project will take. This article is a deep dive on why projects always overrun. It provides a framework to improve future estimation, backing it up with a combination of anecdotal evidence and visualization. The key takeaway is that even a detailed estimate of the work required for a project can still miss significant “invisible” work. Some strategies and thoughts are provided to help not miss estimating such invisible work over time.
This article posits that artificial intelligence (AI) technology is highly capable of replacing executive roles in companies, especially considering its ability to learn from vast historical data. These roles consist of measurable tasks such as capital allocation, creation and execution of market strategy, and candidate selection more effectively than humans. However, the likelihood of this happening is low due to the power dynamics in businesses, with executives being the decision-makers about automation, thus protecting their roles.
TypeScript 5.1 has been released. It has a host of new features, such as snippet completions for @param JSDoc tags, easier implicit returns for undefined-returning functions, namescaped JSX attributes, and more.
Threestudio is a unified framework for 3D content creation from text prompts, single images, and few-shot images, by lifting 2D text-to-image generation models.
Wasp is a Rails-like framework for React, Node.js, and Prisma. It features full-stack type safety, a typesafe RPC layer, simple deployment, full-stack authentication, email sending, jobs, and custom API routes. Several examples of apps created using Wasp are available.
This is a comprehensive guide to training, optimizing, and unlocking the power of natural language processors. It explains the process of training LLMs, including data preparation, model architecture design, model training, evaluation, deployment, and monitoring in a production environment. It also dives into strategies to improve the accuracy, generalization ability, and performance of LLMs. It’s a great guide for anyone interested in learning the current state-of-art techniques for using LLMs effectively.
Facebook and Instagram will block all news articles in California if state lawmakers pass a bill meant to funnel money from the tech platforms to media organizations, a Meta spokesman said in response to the bill. The proposed California Journalism Preservation Act would essentially tax the advertising profits platforms make from distributing news articles. Under the measure, some 70% of the money collected from the so-called "usage fee" would support newsrooms throughout the state.
GitHub stars are a common way of judging the maturity, reputation, and popularity of an open-source project. However, the number of stars a project has isn’t necessarily correlated with the actual value of the project. This article dives deep into the market of buying GitHub stars for projects and how to use other value radars to judge the quality of a project.
Someone has found 34 malicious extensions on the Chrome Web Store with a total of 87 million users. However, the same day this article was published, Google started removing the malicious extensions from the store.