Applying Generative AI in high-stakes settings could lead to deadly accidents where it’s unclear who or what should be held responsible or even why the problem occurred. Everyone agrees that humans should make the final call, but that is made harder by technology that acts unpredictably, especially in fast-moving conflict situations. Until we are able to reliably make Generative AI hallucinate less and be more predictable, it should be applied first to more “boring” applications, like AI in Google Docs. This is already happening - for example, Google has started rolling out the “Write with AI” feature for Google Docs and Slides.
Tesla's Cybertruck design, which features sharp edges and a stainless steel finish, will require the manufacturer to achieve a level of accuracy in the single-digit micron range on all parts in order for it to look perfect. The straight edges of the design will amplify every imperfection. Elon Musk compared the approach to the requirements of products like LEGO and soda cans. Tesla has been criticized for its fit and finishes in the past, but it is generally accepted that the quality of its products has improved over the years.
Meta has released Code Llama, a large language model built on top of Llama 2 that specializes in generating and debugging code. It has also released a Python-specific version and another version that can understand instructions in natural language. These models are not interchangeable. Meta claims that Code Llama performs better than other publicly available LLMs on coding tasks, but it did not specifically name which models were tested.
The constant evolution of EV tech has made it challenging to create infrastructure that can support the wide range of technologies available. A group of researchers have created a novel battery charger that can support a vast range of present and future generations of EV battery packs. The charger is compatible with batteries between 120 and 900 volts with an efficiency consistently greater than 94%. It can be used as a portable unit. The researchers have filed a patent and are in the process of commercializing the technology.
Researchers at North Carolina State University have enlisted the help of robots to help search for the ideal perovskite. Perovskites are a type of material with a rhombus-in-a-cube crystal structure that can be used to create highly efficient solar cells. There are a vast number of possible perovskite materials. Predicting and testing the potential properties of all these materials is time-consuming and difficult. The robot combines chemicals in different proportions to form perovskites and then applies them onto a surface for testing. It speeds up the synthesis and characterization of materials by a factor of 14 compared to manual exploration.
pixi as a cross-platform multi-language package manager and workflow tool. It supports Python, C++, and R on Linux, Windows, and macOS. pixi provides a Cargo-like command-line interface that allows developers to install tools per-project or system-wide. A short video demo is available.
Expresso TS is a TypeScript and Node.js framework for building scalable, easy-to-read-and-maintain server-side applications. It streamlines repetitive and time-consuming processes so that developers can focus on the important things. Expresso TS handles configuring logging systems, authentication, connecting to databases, and more.
Tor's Proof-of-Work (PoW) defense for Onion services is designed to prioritize verified traffic as a deterrent against denial-of-service (DoS) attacks. Tor Onion services are vulnerable to DoS attacks as Tor obfuscates IP addresses. PoW defense will start prompting incoming client connections to perform a number of successfully more complex operations when an Onion service is under stress. The Onion service will then prioritize connections based on the effort level demonstrated by the client. This will thwart DoS attacks without compromising user privacy.
Zoom is requiring employees living within 50 miles of an office to report to the office a minimum of two days a week. The company's CEO, Eric Yuan, says that it's difficult for employees to build trust with each other on a computer screen. He claims that it's too difficult to have innovative conversations and debates on Zoom as it makes people too friendly. His comments point towards the company's belief in the ability of its platform.
This article tells the story of how a former engineer at Alameda Research met Sam Bankman-Fried and provides a few anecdotes from their time working at the firm.
Flannery Associates and its parent company, Flannery Holdings, have purchased approximately 140 properties for more than $800 million in Solano Country at premium rates since 2018.
Email apnea, which, according to an informal survey, happens to roughly 80% of people, is a phenomenon where people periodically hold their breath or alter their breathing when checking their emails.
Natural language interfaces can simplify data exploration. But the data needs to be accessible and trustworthy — which is where the semantic layer comes in. Join this AtScale Tech Talk
Meta has released Code Llama, a Llama 2-based large language model for coding. The 7B, 13B, and 34B parameter models are now available. They performed strongly on HumanEval (53.7) and MBPP (56.2). Code Llama models can generate and debug code, are instruction-tuned, and have extremely long context - as much as 100k tokens.
Hugging Face has raised $235 million in a Series D funding round, with participation from Google, Amazon, Nvidia, Intel, AMD, Qualcomm, IBM, Salesforce, and Sound Ventures. The startup is now valued at $4.5 billion.
The White House, in collaboration with tech giants like Google and OpenAI, has initiated the "AI Cyber Challenge" to develop AI that can spot and fix software vulnerabilities. The competition is offering $20 million in prizes. Winning teams will be encouraged to open-source their solutions.
A pretty large and collaborative effort at constructing a benchmark for measuring language models' ability to reason in a legal framework. Watch out for contamination in already trained models.
Storytelling via automatic video generation is an emerging area of research. The tasks of story continuation, story remixing, and others are tough to measure. Hopefully, this great new Google benchmark helps the field make progress.
Most computer vision tools have trouble tracking 3D objects using LiDAR points because of distractions or not noticing long-term movements. MTM-Tracker mixes two methods and works in two stages to solve these problems.
Graphcore released an excellent Pytorch library that makes training in fp8 super easy and stable. A single model wrapper and torch compile is all you need.
Instead of using big and resource-heavy AI models, Prompt2Model lets you describe what you want in simple words and then makes a smaller, efficient AI model for that task. It can make models that work better than some popular big models.
Recent models often struggle with complex vision-language tasks due to limitations in understanding mixed image-text context. The I4 benchmark was introduced to evaluate these tasks. Results showed a flaw in the Visual Prompt Generator's attention. To fix this, researchers developed Cheetah, a model with a unique module and training strategy that excels in understanding complex interleaved instructions. Cheetah achieved top performance on the I4 benchmark.
Human feedback is key to improving language model alignment and overall performance. However, the process of training with Proximal Policy Optimization has some challenges. Recent work showed that you can optimize directly against human preference and bypass the reward model. With Text-based Reinforcement Learning, you can do this on state-of-the-art open models.
An exploration into why the current crop of LLMs are unlikely to lead to AGI and why physics itself will prevent any sudden leaps from where we are now to AGI.
Bankrupt cryptocurrency exchange FTX is looking to stake, sell, and hedge its ~$3 billion in digital assets. FTX wants to hire Mike Novogratz’s Galaxy as an advisor to optimize its sales. The company's goal is to protect the value of the holdings through calculated trading, potentially via weekly sales limits, and ultimately return funds to users in fiat instead of crypto.
Mastercard and crypto exchange Binance will terminate their cryptocurrency card programs in Argentina, Brazil, Colombia, and Bahrain from September 22. The cards enabled payments in traditional currencies using cryptocurrency holdings. The termination will not affect Mastercard's other crypto card initiatives. This move comes amidst Binance facing ongoing regulatory challenges, including a US lawsuit for alleged deceptive practices.
Beam has launched a product called Instant Payouts. Instant Payouts allows users to send crypto from their cold or hot wallets directly to their debit card and receive funds on the same day. Beam charges a 1% fee, similar to Venmo and CashApp. Instant Payouts is live on Ethereum, Solana, Avalanche, and Polygon.
Maple Finance is returning to Solana to launch its Cash Management Solution, which could provide DAOs, corporate treasuries, and high-net-worth individuals access to yield backed by US Treasury Bills. The pool will accept deposits of USDC-spl, targeting the current 1-month US Treasury bill rate minus fees, and will not impose additional costs on lenders. Maple Finance's return is just one of many big news items on Solana this week.
Cryogen is a tool designed to effectively consolidate extracted datasets from Ethereum while boosting query performance and reducing disk space. It does this by consolidating datasets in place and merging parquets into larger files to keep the dataset updated regularly. Caution is advised when using Cryogen with Cryo as it may output duplicate data in the dataset.
The news cycle on Solana really picked up speed this past week. Highlights include a direct integration with Shopify, a new ChatGPT plugin, the return of treasury management solution Maple, private token swaps with Elusiv, $3 million seed fundraising for collectible content company DRiP, and video call token-gating via Brave. Other notable mentions were the rollout of Solana Sign-In and the launch of the first euro stablecoin on Solana.
Roman Storm and Roman Semenov have been charged with allegedly operating an unregistered money transmitting service. The allegations seem unclear as they appear to have functioned as anonymous software providers within FinCEN's guidelines rather than as money transmitters. The case revolves around whether they had "independent control" over funds transmitted. The specifics of whether they controlled the smart contracts remain ambiguous.
Coinbase announced plans to decentralize the Base network, aiming to create a system where no single entity has control. Working alongside the Optimism Collective and OP Labs, Coinbase will implement fault proofs and diverse client software to avoid censorship. Additionally, they will contribute to the OP Stack infrastructure and help form a "security council" for the Superchain to further enhance security and decentralization.
Chainlink’s new CCIP product enables cross-chain trading, lending, and more and has some crossover with LayerZero functionality. This tweet provides a comparison of the two focused on their security models.
This blog explains why current crypto incentive systems, like token rewards and airdrops, are insufficient for mainstream adoption. It instead suggests that web3 products should draw from Web 2.0 loyalty programs, where incentives are based on higher psychological motivators like status and curiosity.
This thread explains why the framing and discussions around the concentration of staking between different protocols is misguided and could be harmful to the space.
CloudNordic, a large Danish cloud provider, has suffered a ransomware attack that encrypted all customer data. The company cannot and will not meet the ransom demands and advises customers to consider their data lost. CloudNordic doesn’t believe that any data was accessed by the attackers.
Cloud credentials stealing campaigns continue to grow. Cloud service credentials are increasingly targeted because teams are finding more ways to profit from them. One actor has expanded its tooling from AWS to target Azure and Google Cloud Platform. This report dives into how the tooling interacts with the cloud providers and what is new with the latest version.
Secureworks researchers witnessed the Smoke Loader botnet dropping a custom Wi-Fi scanning executable to infected systems. The researchers named the malware ‘Whiffy Recon’. It triangulates infected systems' positions using nearby Wi-Fi access points and Google's geolocation API.
Snowflake's popularity makes it an attractive target, requiring vigilance against emerging threats. This post explores using built-in logging to enhance security operations. It covers ten threat scenarios, relevant exposed data, and a script for quick threat hunting.
BitLocker's protection depends on its configuration, so with a discrete Trusted Platform Module and no password, the BitLocker decryption gets bound to the hardware. This article navigates through how to use physical hacking to obtain the decryption key and hack into a BitLocker-protected computer.
This X thread shares strategies for finding Server Side Request Forgery (SSRF) attacks for bug bounties. These include trying different URL schemes, stealing NT LAN Manager hashes with Responder, trying alternative representations of IP addresses, and searching for references to Kubernetes and hitting its API. The author of the thread suggests being aware of a target's technologies and using open redirects to find SSRF vulnerabilities.