Search the archive

INSIDE GOOGLE’S BIG AI SHUFFLE

AGI FROM FIRST PRINCIPLES

RUNNING LLMS ON PERSONAL DEVICES

This paper introduces INT-FP-QSim, an open-source simulator that tests how well large language models and vision transformers can work at different levels of numerical precision.

MAKING PICTURE RESTORATION BETTER AND FASTER

This research introduces the Dynamic Path-Controllable Deep Unfolding Network (DPC-DUN), an innovative model that optimizes image reconstruction in Compressive Sensing (CS), balancing performance and complexity for enhanced efficiency and results.

LEARNING BETTER VISUAL REPRESENTATIONS

This project proposes an algorithm to filter web datasets used for training CLIP in order to learn better visual representations and achieve state-of-art zero-shot accuracy on vision tasks.

DANSWER

Danswer allows you to ask natural language questions against internal documents and get back reliable answers backed by quotes and references from the source material.

ALIBABA CHAT2DB

Alibaba has launched Chat2DB, a versatile general-purpose SQL client and reporting tool for databases that integrates OpenAI's ChatGPT capabilities. This tool aims to revolutionize the way databases are interacted with, making it more accessible and user-friendly.

SYMBOLIC AND STRUCTURE REGRESSION FOR SCIENCE

WE’RE GOING DOWN A VERY DARK PATH WITH AI REGULATION

DAVIDSON ON TAKEOFF SPEEDS

LLAMA 65B ON RASPBERRY PIS? (GITHUB ISSUE)

REPUBLICANS ASK JUSTICE DEPARTMENT, SEC TO INVESTIGATE PROMETHEUM

Republican lawmakers have called for the DOJ and the SEC to investigate Prometheum, a crypto firm, concerning its relationship with Chinese investors. Prometheum's co-CEO, Aaron Kaplan, had testified before a House committee about the company's ability to operate alongside securities laws. Suspicions arose following discrepancies between Kaplan's testimony and Prometheum's SEC filings. The lawmakers fear that Prometheum may have misrepresented its ties with Chinese investors, primarily Shanghai Wanxiang Blockchain, leading to an insinuation of potential securities fraud.

SHOULD BE APPROVED

Crypto asset manager Grayscale has raised criticisms against the SEC for approving a leveraged Bitcoin-based ETF, alleging that the product is actually riskier than Grayscale’s Bitcoin futures products. This comes amidst Grayscale's ongoing lawsuit against the SEC for rejecting its spot Bitcoin ETF application. The company suggests the SEC is discriminating against different companies, and is seeking legal redress, with a judgment expected by year-end.

THIRDWEB LAUNCHES ONE-COMMAND CONTRACT DEPLOYMENT

Thirdweb has rolled out a new way to deploy contracts on EVM chains. This new secure method eliminates hassling with private keys and copy-pasting ABIs, simplifying the process into a single command. Users can select from a range of pre-built smart contracts such as NFT Drops, marketplaces, tokens, staking, and governance. Post-deployment, developers can hop into a contract management dashboard and leverage SDKs designed for the contracts they deployed to start building their full-stack app. Over 300,000 contracts were launched with this feature in June.

BITCOIN TIPPING ON DAMUS

Bitcoin developers have launched Zapple Pay, a new service enabling the exchange of Bitcoin through emojis in the Bitcoin-friendly app Damus. The innovation comes after Apple threatened to remove Damus from its App Store for allowing users to exchange Bitcoin, a practice it equates to selling digital media. Damus' developer, William Casarin, removed this functionality, leaving users with only the ability to exchange profile-level tips. However, some independent developers found this workaround, arguing that Apple will struggle to interfere due to the third-party nature of their service.

DEVELOPER REPORT 2023 MID-YEAR UPDATE

According to Electric Capital’s Developer Report, the number of active open source crypto developers has declined by 22% over the past year, largely due to newcomers leaving the industry. However, the number of developers working in crypto for over a year has remained steady and has accounted for most of the space’s activity, contributing more than 80% of code commits. Despite the overall drop, the number of active developers is higher than the pre-peak period in 2021. Ecosystems such as Osmosis, Sui, Aptos, TON, Optimism, and Aztec Protocol have reported year-over-year growth.

SAFE MODULAR SMART ACCOUNT ARCHITECTURE – EXPLAINED

Safe is propelling innovation in modular Smart Account infrastructure, enabling developers to construct a variety of applications and wallets. Its framework emphasizes four primary facets: Plugins, Hooks, Function Handlers, and Signature Verifiers. The structure allows for the creation of unique functionalities while maintaining secure operations. The architecture focuses on Safe Account, Singleton contracts, Factories, and Integration Interfaces, with considerations around cost efficiency, upgradability, and augmented functionality. Safe's structure incorporates core Ethereum principles like signature thresholds to ensure security and extend its capacities to allow developers to build robust smart wallets.

ETHEREUM’S NEW DATA ECONOMY

Ethereum’s next major upgrade, called "Dencun," will focus on EIP-4844. The upgrade aims to reduce transaction costs on mainnet and improve the way L2s encode data on mainnet, significantly affecting the data economy of the platform. The data associated with the upgrade will be temporary, persisting for only 30 days. This new approach has implications for applications using on-chain data, especially NFTs, and raises questions about future data storage and availability.

ABOUT THE COMPANY DOXXING ITS USERS

Arkham Intelligence is launching a marketplace and token that allows users to pay for on-chain sleuthing. The platform, Arkham Intel Exchange, uses a bounty system where users post a bounty and whoever fulfills it first gets paid in Arkham’s native coin, ARKM. Alternatively, users can auction off their independent findings if they are of interest to potential buyers. In addition to its token launch, Arkham gained some attention yesterday as news came out that they had been doxxing users’ email addresses in their referral links. While the links looked harmless, they were actually the users’ Base64 encoded email addresses, which tied their emails to their Twitter accounts when they unwittingly shared them.

A POSTMORTEM: MY EXPERIENCE AS A DEFI STARTUP CO-FOUNDER

This thread lays out the experience of an individual who co-founded a crypto startup, sharing the insights he gained on crypto, entrepreneurship, and ethics.

BITCOIN’S NEXT TARGET: A LUNA LEFTOVER

In May 2022, the LUNA collapse led to significant BTC sell-offs and overall market chaos, causing a notable CME (Chicago Mercantile Exchange) gap of around $35,100. Today, BTC's potential to fill this gap depends on ETF approvals and a declining U.S. Dollar Index.

ORDINALS INSCRIPTIONS SURGE FOLLOWING BRC-69 LAUNCH

Following the launch of BRC-69, a token standard reducing fees and introducing recursion, Ordinals are picking up in activity once again.

ENFORCEMENT

Crypto scammers are increasingly impersonating regulators and law enforcement officials to defraud victims. Despite intense government scrutiny and efforts to curb such fraud, these impersonation scams have still been effective and difficult to prosecute due to their international and mutating nature.

THREADS ALREADY HAS OVER 95 MILLION POSTS

Threads had more than 30 million sign-ups and 95 million posts in its first 24 hours. It reached the top free apps on the App Store. There are still a lot of features missing from the app, such as DMs or a reverse chronological following feed. The app is also missing a planned integration with the decentralized ActivityPub protocol.

APPLE IS INVENTING A REVOLUTIONARY CAR AUDIO SYSTEM FOR APPLE CAR

Apple is inventing a whole new stereo system for the Apple Car. The company recently filed five patent applications about audio in the car. It uses the phrase 'enclosed environment' as a euphemism for the car, but the way the term is used strongly suggests that it can't refer to anything else but a vehicle. This article provides an overview of the five patents along with images from the applications.

INDIA TO LAUNCH CHANDRAYAAN-3 MOON LANDER MISSION ON JULY 14

India will launch the Chandrayaan-3 mission on July 14, almost four years after the crash of its previous iteration. The mission involves sending a lander, propulsion module, and rover to the lunar surface to conduct on-site scientific experiments. It should land on the Moon in August. If successful, India will be the fourth country in the world to achieve a soft landing on the Moon.

TEMPERATURE

A new type of gel allows 3D printers to print heat-sensitive electronics. It enables the 3D printing of metal objects at room temperature. The gel temporarily turns a metal alloy into a liquid-metal state - it becomes rigid and highly electrically conductive after it is left to dry. It can create objects that can change shape in a predictable fashion if heat is applied while it is drying, allowing for the production of items with complex 3D shapes that are easier to print as a flat pattern.

COMPLETIONS API

All paying OpenAI API customers now have access to GPT-4. The GPT-3.5 Turbo, DALL·E, and Whisper APIs are also now generally available. Models using the Completions API are set to be deprecated in favor of models using the Chat Completions API, which often provides better results. Older embeddings models will also soon be deprecated as the current recommended model, text-embedding-ada-002, accounts for 99.9% of all embedding API usage. The Edits API will be deprecated in January. Details about which exact models will be deprecated are available in the article.

TYSON

TySON (TypeScript Object Notation) is an embeddable configuration language that generates JSON. It can create JSON with comments, types, and basic logic using TypeScript syntax. The project aims to make it possible for all major programming languages to read configuration written in TypeScript using native libraries.

TWITTER IS THREATENING TO SUE META OVER THREADS

Twitter has accused Meta of poaching former employees to create Threads. It is threatening to sue Meta to enforce its intellectual property rights. Twitter's lawyers allege that Meta hired and assigned former Twitter employees to develop Threads with the specific intent that they use Twitter's trade secrets and other intellectual property to accelerate development. A Meta spokesperson responded by saying that Twitter's accusations are baseless as no one on the Threads engineering team is a former Twitter employee.

MY SMALL, NO NAME COMPANY HAS COMPLETELY LOST ITS MIND WITH AI

Some people really think that AI is a magical thing that can fix all their problems. This article gives an example of a company that is using AI to generate JIRA tickets, code, and HR documents without any care about quality or security or checks for legality. The entire process has become a failed experiment that the company's engineers have had to clean up, but the company's decision-makers continue to push the technology to please the CEO. Examples of no name companies acting like this serve as an indication of what the future might hold.

TSMC

Google is sticking with Samsung for next year's Tensor chip design and manufacturing, but it has been replacing Samsung components with its own IP every generation.

THE SECRET LIFE OF .WELL-KNOWN

The .well-known directory was intended to be a generic and extendable version of robots.txt.

TWINSPARK

TwinSpark is a declarative enhancement for HTML that transfers a lot of the common logic from JavaScript into a few declarative HTML attributes to allow developers to make interactive sites with little JavaScript and more manageable code.

KEEPING UP WITH BIG SHIFTS IN TECH

This article features interviews with several developers about their views on how to keep up with the fast-paced changes of the tech world.

META WILL NUKE YOUR INSTAGRAM ACCOUNT IF YOU DELETE THREADS PROFILE

Threads profiles can be deactivated at any time, but deleting a Threads account requires deleting the linked Instagram account.

NEXT WEEK

Code Interpreter lets ChatGPT run code to analyze data, create charts, edit files, perform math, and more.

OPENAI MAKES GPT-4 GENERALLY AVAILABLE

Get out from under GPT-3.5 and start using GPT-4, which is generally available starting today.

CHATGPT DROPS ABOUT 10% IN TRAFFIC AS THE NOVELTY WEARS OFF

Worldwide desktop and mobile web traffic to the ChatGPT website dropped 9.7% from May to June, according to preliminary estimates. In the U.S., the month-over-month decline was 10.3%. Worldwide unique visitors to ChatGPT’s website dropped 5.7%. The amount of time visitors spent on the website was also down 8.5%. ChatGPT still attracts more worldwide visitors than Bing, Microsoft’s search engine.

DEVELOPMENT

LlamaIndex has launched version 0.7.0 with the aim of improving the library's modularity and customizability. The new version includes standalone LLM abstractions for OpenAI, HuggingFace, and PaLM, allows users to use the response synthesis module independently, and offers enhanced metadata management for Document/Node objects. The update was designed to enhance LLM application development and offer developers more control over their workflows.

TEXT-TO-IMAGE EDITING THROUGH FEATURE CORRESPONDENCE

This repo introduces DragonDiffusion, a novel method that enhances the editing precision of text-to-image models. Utilizing feature correspondence within diffusion models, DragonDiffusion allows various detailed image alterations, such as moving and resizing objects, replacing appearances, and dragging content, all while maintaining the image's original consistency and without the need for additional model fine-tuning.

MODELS GENERATING TRAINING DATA: HUGE WIN OR FAKE WIN?

There has been buzz the past few days around a number of works suggesting that models generating training data is a new unlock for performance. On the other hand, the community has pushed back, saying that the data processing inequality suggests this is impossible. What if the truth is somewhere in between and generating training data acts as a sort of filter for sampling from the distribution you actually care about?

ENHANCING PROBLEM-SOLVING ABILITIES OF LLMS

This technical report details the creation of FLACUNA, a language model that improves problem-solving capabilities by fine-tuning VICUNA, another large language model, using a custom instruction dataset called FLANMINI.

IMAGE SEGMENTATION WITH HIERARCHICAL, OPEN-VOCABULARY LEARNING

This paper presents HIPIE, a novel approach to image segmentation that accounts for complex scenes by incorporating different semantic levels into its learning process. It outperforms existing models across a range of image comprehension tasks, offering a unified solution for segmentation tasks in diverse contexts.