AI News & Launches

Brian Merchant's article 'AI really is smoke and mirrors' critiques the generative AI industry, expressing skepticism about the lofty promises made by tech executives. The article highlights the hype-driven nature of AI developments, the underwhelming performance of AI services, and compares modern AI's situation with historical illusions created by 'smoke and mirrors'. It discusses the industry's challenges, such as overly optimistic valuations and the reality of AI's capabilities, suggesting that generative AI could face a moment of reckoning once the facade of its potential wears off.

Intel unveils its new Gaudi 3 AI accelerator chip featuring a groundbreaking 128GB HBM2e memory. Designed for AI inference and training, it's due for volume production in late 2024. Boasting up to 1.835PFLOPS of FP8 compute with 64 tensor cores and 8 matrix math engines, this 5nm chip represents a significant leap over its predecessor, the Gaudi 2.

Apple has reportedly acquired the French startup Datakalab, specializing in AI compression and computer vision technology. The Paris-based company was known for developing efficient deep learning algorithms capable of running on device, and had worked with the French government and companies like Disney on various projects. This move is seen as a step towards enhancing Apple's AI and Vision Pro capabilities.

Jonathan Bartlett explores the fundamental differences between AI and human cognitive abilities, highlighting that computers are theorem generators bounded by axioms they can't establish, whereas humans are axiom generators capable of establishing and processing foundational truths.

Simon Willison discusses the practical applications of Large Language Models (LLMs) for data journalism, through numerous demos at the Story Discovery at Scale conference. He highlighted the recent flurry of AI technology progress using real-world tools that journalists can utilize to analyze data, providing solutions for common challenges like scraping, enriching data, and semantic search.

Chat with Meta's Llama3 allows users to interact with the latest AI chat model, Llama3, developed by Meta. The platform offers unlimited, free access to users for engaging with the AI which can answer questions, generate code, and assist in various tasks.

The recent release of Meta's Llama-3 AI model poses significant competition to proprietary AI models, matching and potentially surpassing existing models like GPT-4. With Meta’s financial and computational resources, Llama-3's impact extends to both industry giants and AI product startups, intensifying the AI race.

A study titled 'Lossless Acceleration of Large Language Model via Adaptive N-gram Parallel Decoding' introduces a novel method for accelerating inference in Large Language Models (LLMs) without loss of output integrity. The proposed Adaptive N-gram Parallel Decoding (ANPD) uses a dual-phase approach for rapid drafting and verification of token generation, achieving up to 3.67x speed improvements in tests.

This article outlines crucial steps for preparing unstructured data for LLMs, emphasizing the need for effective ingestion and preprocessing. It discusses the transformation, cleaning, chunking, summarizing, and embedding generation processes for RAG-ready data, tailored specifically for enterprise-grade applications.

The opinion piece at Humanise.org discusses the impact of artificial intelligence on architectural heritage, positing a future where AI might strip historical buildings of their unique character and 'soul'. It argues that while AI can help in preserving the physical aspects of these structures, the intangible qualities that endear them to people's hearts may not be captured by technology alone.

AllMind Investments specializes in asset management and financial advisory services, aiming to provide superior returns and wealth growth opportunities for investors.

Stylar AI is a revolutionary graphic design and image generation tool powered by AI, providing extensive control over composition, predefined styles for simplified designing, and multiple features like drag-and-drop composition, one-click enhancements, high-resolution exports, and intuitive controls designed for all skill levels. Sign up for 200 free monthly credits during Beta.

Skyla Chat offers a powerful AI chatbot for Shopify merchants to enhance customer service. It is highly customizable, simple to install, and built for best in class performance, providing 24/7 support and improving overall customer experiences.

Chatbot Arena is a platform that allows users to compare different AI chatbot builders to find the most suitable one for their needs. It features comprehensive offerings, pricing details, and community trust to help make informed decisions.

Meta integrates a new AI tool into Instagram and Facebook, which cannot be turned off. This AI is designed to help with tasks like recommending restaurants or explaining subjects for students. While group chats can disable it, its general presence on platforms remains constant. Meta aims to use feedback to improve the tool and its policies.

OpenAI's GPT-4 has been shown to autonomously exploit real security vulnerabilities by interpreting CVE advisories, outperforming other models and vulnerabilty scanners.

Microsoft has developed VASA-1, an advanced deepfake AI framework capable of creating convincing videos from a single image and an audio sample. Despite its potential, they won't release it due to the high risk of misuse, aligning with growing concerns and regulatory efforts surrounding AI impersonation and deepfakes.

The 'code-interpreter' repository on GitHub, managed by e2b-dev, is a public project offering a Python & JS SDK for building custom code interpreters. It's integrated with E2B - Cloud Runtime for AI Agents and enables AI frameworks to share context across code executions, mainly for AI models like GPT-3.5. It supports content streaming and is designed to run on serverless and edge functions.

Parny offers an interactive on-call management and alerting service integrating with over 40 monitoring and cloud services. It's designed for alert management, engagement like social media interactions, and insightful analytics, including DORA metrics.

Grimo combines the idea of Obsidian with Github & Quora to create a platform for knowledge management. Users can query to fork knowledge into their workspace, benefiting from seamless information intake from sources like YouTube, Podcasts, and AI Apps without needing extra plugins.

Vidnoz AI is an AI video generator that offers a free AI video creation tool, with 600+ realistic AI avatars, 700+ video templates, and leading AI voice cloning technology. Users can access free AI tools for face swapping, enhancing videos, and voice changing. The platform allows for rapid creation of professional-looking videos, suitable for various use cases such as e-learning, marketing, communication, and more.

Stevie AI offers an AI-driven platform tailored for startups to enhance their SEO and online visibility. Without the need for SEO expertise, users can improve their website through simple steps with Stevie's guidance on keywords, content creation, site audits, and performance analytics. Plans are affordable with free trials available.

Ampere Computing is targeting AI inference workloads with the upcoming release of its 256-core CPU, leveraging high-performance cores and memory bandwidth improvements. This processor architecture aims to compete with GPUs, typically preferred for AI tasks, by providing cost-effective inference capabilities integrated into the CPU socket. Ampere's strategy incorporates advanced manufacturing processes and a chiplet design, potentially changing the server CPU market dynamics for AI applications.

Chandler K introduces the basics of Google AI Studio and the Gemini 1.5 Pro model, highlighting its multimodal features and how it compares with the Gemini API and other tools like OpenAI's Playground. Key features include choosing from various models, customizing 'creativity level' and 'stop sequence', and ensuring safe content generation. He also explains the different modes, such as Chat Prompt and Structured Prompt, and the ability to interact with media for unique use cases. The article serves as a detailed beginners' guide to leveraging these tools for innovative AI development.

Intel revealed 'Hala Point,' the largest AI neuromorphic computer emulating the human brain. Hala Point, powered by Intel's Loihi 2 processors, boasts 50 times the speed and 1/100th the energy use of traditional CPUs and GPUs. It's designed for AI research and initial deployment at Sandia National Labs for computing and science issues.

Supersimple's early access to GPT-4 fine-tuning shows a performance improvement of over 50% compared to GPT-3.5 for their data question-answering use case, but with higher latency and cost. GPT-4 models continue to face diminishing returns and challenges with broad queries. Supersimple advocates for a mix of models and heuristics to enhance performance.

Paces (YC S22), a company focused on accelerating green development through big data and AI, is currently hiring experienced engineers in New York City. Their objective is to make a significant positive impact on the climate. Interested candidates can explore opportunities on the Paces career page.

Open Agent Studio provides a powerful no-code agent editor for building robust, future-proof RPA tools without coding.

The website provides information about DuckDuckGo, a search engine that emphasizes protecting searchers' privacy and avoiding the filter bubble of personalized search results.

Lois Beckett reports on a visit to CaliExpress, a 'fully autonomous' restaurant in Los Angeles that uses AI-powered robots, such as Flippy, to cook fast-food like burgers and fries. Amidst labor cost debates and wage increases, CaliExpress demonstrates a potential future where humans collaborate more with technology than execute menial tasks, yet the experience leaves Beckett pondering the quality and implications of such automation.

Discord's recent blog post outlines their rigorous process of building features with Generative AI, dividing the journey into stages such as use case identification, product requirements definition, prototyping, and full-scale deployment. The post emphasizes the potential of AI in enhancing user experiences and solving large-scale problems and discusses the complex trade-offs involved in integrating LLMs into applications.

This post discusses the current state of AI in assisting with hand-drawn animation inbetweening, evaluating two academic papers and a commercial tool. Traditional inbetweening, cleanup, and coloring are considered, with focus on productivity enhancements via AI. Limitations are acknowledged, particularly with handling occlusion and rotation, but advancements in 3D motion reconstruction and feature matching techniques show promise for future improvements.

AIFlash.Cards is a cutting-edge tool that allows users to create personalized flashcards from text or files, aiding in faster learning and better retention. It offers features such as generating flashcards from topics, contexts, links, and PDFs, and supports 100 languages. The service boasts spaced repetition, cross-platform access, history tracking, and community engagement. It has free and premium plans.

Pietra Studio provides a platform for dreamers and entrepreneurs to design and manufacture products using AI. Over 250,000 brands use Pietra to save time and money, creating designs and variations in seconds. The service offers collaboration tools, thousands of templates, and a design studio to streamline the creative process.

HuggingChat is a platform that provides access to the community's leading AI chat models, making them available for general usage while also raising awareness about potential issues like biased data generation and misinformation.

The US Air Force and DARPA have showcased AI software's ability to fly a modified F-16 fighter jet, achieving a milestone in machine learning and autonomous combat aviation by dogfighting against human pilots. The breakthrough, considered a transformative moment for combat aviation, may impact both military and commercial aviation.

Tiger is an open-source, community-driven project that serves as an integrated tool ecosystem for LLM Agent Revolution. It provides a platform, modeled after Neuralink, enabling AI agents to interact with computer interfaces through function calls. Tiger supports a plethora of functionalities, including code execution, search, system info, knowledge management, and communication, mainly through Telegram integration. It is MIT licensed, aiming to build a robust community around AI tool support.

The paper titled 'Formal Mechanised Semantics of CHERI C: Capabilities, Undefined Behaviour, and Provenance' presents a comprehensive definition of CHERI C semantics and discusses key design questions related to capabilities, provenance, and undefined behaviour. This work is foundational for CHERI C, aiding those porting code, compiler implementers, and future verification.

Vapi is a platform designed to enable developers to create voice AI applications that are as easy to integrate as any other API. It is aimed at simulating natural, human-like conversations with optimizations for low latency and real-time interactions. Vapi provides an orchestration layer over various technologies, including Speech-to-Text and Text-to-Speech, to make voice AIs talk like a person.

Gentoo Linux project has unanimously decided to ban AI-generated contributions due to concerns over copyrights, quality, and ethics. Górny's proposal prohibits use of AI tools like ChatGPT and GitHub Copilot for creating content within the Gentoo project.