Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Gemini Pro 1.5 is an enormous upgrade for the Gemini series. The model has a 1,000,000 token context size, much larger than Claude 2.1's 200,000 and gpt-4-turbo's 128,000 token context sizes. While the model can still miss things and hallucinate incorrect details, it is able to process and extract text information from short videos. This article contains an example where Gemini Pro 1.5 is used to extract book names from a short video.
Apple's PQ3 cryptographic protocol purports to offer more robust encryption and defense against quantum computing attacks.
A summary of Nvidia's fourth fiscal quarter earnings.
AI is the most Douglas Adams of all technologies - when technology becomes absurd, we must respond with absurd inventions.
The top three phones of 2023 were all the iPhone 14 models, with the iPhone 13 taking the fourth spot before the iPhone 15 models in spots 5, 6, and 7.
A collection of questions about the AI industry and where it is headed.
Retell AI turns LLM agents into human-like voice agents that can handle tasks like making appointments with customers.
Sponsored
Have you been looking for a better way to ground your LLM app to avoid hallucination? In this free 4-hour course [5], you will learn how to:
Google has released the weights for its Gemma 2B and 7B parameter models, available via HuggingFace. The models are decoder-only Transformers trained on 2T and 6T tokens. They substantially outperform Llama 2 on a wide range of benchmarks and come in base and instruction-tuned versions.
The language modeling group at Stanford released a widely used benchmark called Holistic Evaluation of Language Models (HELM). They also released a version for instruction following called HELM-Instruct. It is multidimensional, open-ended, and absolute.
Reddit has entered a licensing agreement worth approximately $60 million annually with an unnamed major AI company, granting the company access to its vast user-generated content.
Predibase released 25 narrow task-specific LoRAs that each outperform GPT-4 on specific tasks. These task-specific LoRAs are built onto the Mistral 7B model. This is another piece of evidence to support the ability of local models to perform exceptionally well on exact tasks.
ChartX is a new tool for testing multi-modal large language models (MLLMs) on their ability to interpret and reason with visual charts.
This repo challenges the effectiveness of combining supervised fine-tuning and reinforcement learning in training. Using a more advanced model like GPT-4 for initial training can outperform the more complex two-step method.
Meta has released the MMCSG dataset, which has thousands of two-sided conversations recorded via the Aria glasses, to encourage research in various social audio and visual tasks.
Lorax is an inference server that supports hot-swapping LoRAs onto a single base model. This means a dramatic decrease in RAM while supporting a wide range of model tunes.
This project introduces GTBench, a platform for testing LLMs like GPT-4 in various game scenarios to discover their strengths and weaknesses in strategic reasoning. While these models show promise in games based on probability, they struggle in deterministic games and vary in performance based on their design and training.
A diffusion model that can generate multiple views of an object and then stitch them together into a compelling 3D version.
AI is going to automate parts of the workforce. The reaction so far has been two-sided – will this efficiency mean the same humans can do more or will fewer humans do the same tasks? This article explores comparisons between the impact of tech on manufacturing and agriculture and what we’re seeing today with the modern knowledge worker.
ChatterBox is a vision-language model that excels in multimodal dialogues, specifically in the newly established task of multimodal multi-round referring and grounding.
Sponsored
For the next 12 hours, try Dollar Flight Club for just $1 [18] to make it happen and save up to $500 bucks on your next trip. Join 1 million members saving on adventures this year [18]. Make
Apple is working on an updated version of Xcode that will include an AI tool for generating code.
Create 100+ high-performing ad variations with AI.
CrewAI offers a library for building and orchestrating AI agents that leverage LangChain and Replit. It provides a modular setup with agents, tasks, tools, and crews for easy integration into various projects. Debugging is simplified through LangSmith, which enhances performance insights into non-deterministic LLM calls.
Hack VC has secured $150 million in funding for a web3-focused fund. The firm was intending to raise less, but the round ended up being oversubscribed, reflecting shifting sentiment amongst crypto investors. Among other things, Hack VC aims to invest in tokenized real-world assets and modular infrastructure. Other firms, like Lightspeed, Faction, and RW3 Ventures have also raised impressive funds.
Texas-based crypto firm Lejilex and lobby group Crypto Freedom Alliance of Texas have sued the SEC, accusing it of overstepping its authority in regulating digital assets. They contend that the SEC lacks clear jurisdiction over the crypto industry and have asked the court to recognize that digital assets on exchanges not be considered securities. The entities also seek to prevent the SEC from suing its members for listing pre-existing tokens.
The Modular Liquidity Layer, an AMM on Dymension, has launched. It was developed as a settlement and liquidity layer aimed to centralize liquidity for easy access to modular assets and enable smooth routing within the network. It is laying the groundwork for the many RollApps that will launch on Dymension. RollApps will be able to leverage the AMM, making it a modular ecosystem launchpad. DYM will be fundamental for every pair in this liquidity layer.
Cross-chain communication platform Wormhole will use AMD's enterprise-grade FPGA hardware accelerator chips to facilitate secure cross-chain messaging using zero-knowledge proofs. Aimed at fortifying "corridors" between blockchains, the project is also developing secure cross-chain transfer "light client" implementations. AMD will contribute both the chips and its hardware acceleration expertise to increase power for proof generation and verification across networks.
MyShell, an AI creators' platform, has launched a Patron Badge system that allows its AI agents and apps to become investable assets. The system involves users buying into an agent or app and receiving shares of the profits generated from user interactions. The innovative system is designed to not only support the creator economy but also fuel the platform's growth. This move represents MyShell's long-term vision to build a robust ecosystem around AI models.
PhotoFinish is a Virtual Horse Racing game that has been gaining steam lately. It has officially partnered with NBCsports and Churchill Downs. Players can get involved in a number of ways: racing, horse sales, breeding, track ownership, and wagering. This thread provides a brief overview of how you can take part in each of these activities and some examples of how much players have earned in the game recently.
Yuga Labs' co-founder is reassuming his role as CEO, prioritizing a focus on crypto natives. BAYC is transitioning to a detachable Yuga subsidiary, BAYC LLC, to improve its focus and agility. Otherside, a community platform, is now aiming to become the default space for community engagement. Yuga’s gaming strategy will aim for user-friendly and crypto-native mechanics.
The NIL Foundation is making strides in scaling Ethereum, focusing on zkSharding and proof markets. This thread outlines the key benefits of using NIL's zkRollups, including nearly 60k TPS scalability, security, liquidity, and enhanced functionality with Ethereum's staking mechanism. The system essentially works on zkLLVM, a Placeholder Proof System, Crypto3, and a Decentralized proof market.
Beam is now live on Base, enabling instant bank transfers of USDC or ETH from the Base network via Coinbase Wallet or others and simplifying crypto cashouts without needing cross-chain bridges or exchanges.
Tom Lee of Fundstrat predicts that Bitcoin could hit $150,000 in 2024, driven by new ETFs, the halving event, and the anticipated easing of monetary policy.
Circle discontinued USDC support on Tron earlier this week due to issues with compliance and risk management amidst allegations of Tron's involvement in illicit activities.
Montenegro's High Court rejected a request to extradite Terraform Labs' co-founder Do Kwon to South Korea, opting instead for his extradition to the US to face fraud charges.
Despite much competition, Photoshop is still a huge design tool in 2024. The Knoll brothers developed the first version of Photoshop on February 19, 1990. It was immediately popular, and Adobe bought the rights for $34.5M in 1995. After over three decades, with AI taking the world by storm, Adobe has integrated generative AI software into its suite of applications. The company remains a design powerhouse in the industry.
Google is introducing a "more modern look and feel" to its sign-in page that features a light Material Design makeover with horizontal alignment for better screen compatibility. The sign-in process, including password prompts, remains the same. The rollout will be gradual, with no control granted to Workspace admins or personal Google account users.
WhatsApp has introduced four new text formatting options — bulleted and numbered lists, block quotes, and inline code — to help users better organize and format their messages. The new feature is supported in Channels. Users can employ markdown-style syntax within messages to utilize these formatting options efficiently. The new formatting options are now available across Android, iOS, Web, and Mac platforms.
Whether you are developing a personal brand, start-up, or refining an already existing brand, the following 5 UX phases help to structure a methodology that sets a clear path to defining your brand voice with measurable outcomes: Know and recognize your audience, state your business values, define your brand’s tone, gather feedback, and trust and entitle your team to ensure brand voice adherence.
Users rarely read on the web - they usually scan instead, which could be more efficient for them, but very detrimental to your business. This article explores how users consume text online and how to prevent harmful patterns. To mitigate scanning, designers can use headings and subheadings, visually group small chunks of content, and alternate sizes, spacing, and patterns.
Planning, research, and conceptualization are essential to creating an icon, but they must be done before crafting to ensure the right design decisions occur. Establishing clear goals, setting the tone, and organizing icons are essential steps in an icon project.
Galileo AI is a UI generation platform that allows designers to explore prompts and designs from other designers and generate UI from text prompts or uploaded images.
Generate color palettes from text prompts, apply and preview them on UI, and explore and switch between the palettes to find the right one.
Calculate how many people you need for your next research round based on your project's scope, participant characteristics, researcher expertise, and methodology.