Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Sponsored
How do you broaden access to data and AI? The Semantic Lakehouse is currently gaining traction as a way to increase data engineering efficiency while maintaining a tool-agnostic architecture. Learn how it works in this upcoming webinar [12] featuring Databricks architect Pulkit Chadha and AtScale [12] CTO Dave Mariani. Save your spot,
Reranking documents is key for retrieval but hard to do in practice. This library provides tools for reranking documents. It can be integrated into your retrieval pipeline.
Skyvern automates browser-based workflows using LLMs and computer vision. It provides a simple API endpoint to fully automate manual workflows, replacing brittle or unreliable automation solutions.
This project introduces ViT-CoMer, a new backbone architecture enhancing Vision Transformers (ViT) for dense prediction tasks without the need for pre-training.
NVIDIA co-founder Curtis Priem has donated $275 million to Rensselaer Polytechnic Institute (RPI), impacting its technological advancements and allowing it to house an IBM Quantum System One computer. He gave away his NVIDIA shares after the IPO, valuing meaningful contributions over wealth retention. Priem's philanthropy has been pivotal in enhancing RPI's academic and research infrastructure.
Arvind Jain, CEO of Glean, discussed the challenges and strategies for enterprise AI startups at a Lightspeed meetup, emphasizing the importance of delivering true business value, persistence in hiring, and prioritizing product quality over cost and speed. Jain also highlighted the slower adoption of generative AI tools in enterprises due to privacy and security concerns. Glean aspires to be a ubiquitous enterprise AI platform, embedding deeply within organizational workflows to fundamentally change how people work.
Anthropic's new AI model, Claude 3, stands out for its 'warmth,' making it a robust partner in creative writing tasks. Claude 3 is described as more human-feeling and naturalistic, crossing the threshold from good thought to deep thought made enjoyable. Despite technical benchmarks not fully capturing this nuance, Claude 3 is poised to revolutionize how we interact with AI in creative processes.
Assort Health has announced a $3.5 million funding round to scale its generative AI solution designed specifically for healthcare call centers.
Regulating AI like a physical object is unfeasible due to its nature as a process integral to numerous applications across the economy.
The Full Self-Driving (FSD) Beta v12.3 update, which Elon Musk has described as a "big release", uses end-to-end neural nets for vehicle controls, a shift from programmer-coded controls.
Sponsored
When ideas are clearly communicated, innovation can flourish. Narratize [5] uses genAI to help innovators gain buy-in for groundbreaking ideas — and empowers marketers to transform technical knowledge into value-driving collateral.
Together AI, a compute provider and AI research group, has raised additional funding in a round led by Salesforce Ventures and other top VCs. The company is growing 3x month over month.
Google DeepMind's SIMA is a generalist AI agent for 3D games that follows natural-language instructions across various video game environments, marking a shift towards creating versatile, instructable AI systems.
Applied Intuition, a leader in autonomous vehicle software, raised $250 million in Series E funding. This values the company at $6 billion, a 67% increase from its previous valuation. Its generative AI tech is used in developing software for automotive, trucking, and other industries, leading to profitable triple-digit growth and partnerships with major manufacturers.
Teams from Meta and IBM have combined to achieve 57% MFU, using massive A100 and H100 clusters to quickly train powerful models in parallel.
This research introduces an innovative approach to video generation using state-space models (SSMs), overcoming the memory constraints of traditional attention-based diffusion models by focusing on linear memory consumption. SSMs enable the creation of longer video sequences with competitive quality, as demonstrated in tests with the UCF101 and MineRL Navigate datasets.
MoAI is a novel large language and vision model that enhances current models by integrating auxiliary visual information from specialized computer vision tasks.
Sponsored
Keep up with the latest breakthroughs in AI, security, data, and more at Google Cloud Next ‘24:
→ Learn about implementing enteprise genAI [14] and AI for media
→ Discover the 3 secrets ingredients to accelerating value from
Save your spot at the special rate of $1,599 USD by using code COMETONEXT24 by 11:59 PM PT on March 31, 2024. Register now ↗️
Building with LLMs requires, for the most part, building systems around close models. These systems will include things like prompt tracking, cost estimations, caching repeat requests, debugging, and privacy. superopenai is a minimal tool that provides functionality for a wide range of these features.
TripoSR is a method that uses a forward pass model to reconstruct 3D models from 2D images. It runs extremely quickly and is close to state-of-the-art for the task.
SemCity revolutionizes 3D scene generation by focusing on real-world outdoor environments, a challenge often overlooked due to the complexity and emptiness of outdoor data.
AI startups face unique challenges compared to past tech revolutions: they're directly competing against well-funded incumbents with mature products, existing customer data, and top talent, making it tougher to find strategies that yield a competitive edge in the AI market. Traditional startup advantages have diminished in the AI space, requiring novel approaches to innovation and market entry. Without adapting to these realities, many AI startups risk failure despite initial hype and excitement.
Recent tests reveal that advanced AI models like ChatGPT-4 and Google's "Gemini Advanced" struggle with visual IQ tests, scoring below thresholds for measurable IQ. While ChatGPT-4 shows marginal pattern recognition skills, it makes visual misidentifications and logical errors, suggesting a significant gap compared to human intelligence. These findings imply that we might still be some distance away from achieving generally intelligent AI systems.
This article urges UX designers to innovate beyond Quick Actions and Multi-Turn patterns in GenAI interfaces, addressing the need for more intuitive, adaptable designs that accommodate the evolving nature of AI. Emphasizing user empowerment and contextual understanding, it highlights promising alternative UX patterns like drag-and-drop, voice interactions, and AR overlays. The exploration of these alternatives aims to enhance user satisfaction and foster meaningful interactions with AI systems.
Sponsored
Apache Airflow is playing a key role in the AI/ML initiatives of modern enterprises. Find out how to start delivering production-ready AI with Airflow in this free ebook by Astronomer [24].
OpenAI plans to publicly release its text-to-video model, Sora, later this year after conducting thorough testing for vulnerabilities and setting up safety guardrails.
Google's March core update to its ranking algorithms will show fewer results that feel like they were created for search engines instead of people.
Andreessen Horowitz's latest consumer report on the top 100 Gen AI consumer apps found that over 40% of the top web products are new, having entered the top 50 in the last 6 months.
Sponsored
Don't miss your chance to experience the latest AI breakthroughs in-person at NVIDIA GTC [4], the #1 AI conference [4]. NVIDIA CEO Jensen Huang will share the latest AI advances that are shaping our future, live at the SAP center in San Jose. With 900+ inspiring sessions [4], 300+ exhibits, and tons of unique networking events, GTC delivers something for every technical level and interest area.
Cognition announced a new system called Devin that scores 14% on SWE-Bench, a challenging benchmark that tests AI’s abilities to write code. GPT-4 scored 1.7%. This model shows strong in-context learning abilities.
This blog post from Meta outlines the infrastructure being used to train Llama 3. It talks through storage, networking, Pytorch, NCCL, and other improvements. This will lay the foundation for Meta's H100s coming online throughout the rest of this year.
Physical Intelligence (Pi) has emerged from stealth with $70 million in seed funding. The company, founded by a team of renowned robotics and AI experts, aims to develop foundation models and learning algorithms that can power a wide range of robots and physically-actuated devices.
You can steal portions of closed language models such as the embeddings layer just by using their public APIs. This can be done for a modest budget of less than $2,000.
A sequence prediction model for DNA built on the Transformer competitor Mamba. It is extremely efficient and powerful for a small model.
This project introduces a novel approach that enhances 3D object generation, producing high-quality, detailed objects from a single image.
Sponsored
DataGPT is your on-demand, always available data analyst. From simple reporting to cohort and trend analysis [11], DataGPT gives reliable, thorough, and accurate answers to natural language questions like “why did our conversions drop?” 🆕 Use Dynamic Benchmarking [11] to compare two segments in your data, including from different starting dates.
Transformer Debugger is a tool developed by OpenAI's Superalignment team to support investigations into specific behaviors of small language models. The tool combines automated interpretability techniques with sparse autoencoders.
VideoMamba is a solution that addresses the complexities of video understanding by efficiently managing local redundancy and global dependencies.
This project presents a way to improve the efficiency of Large Vision-Language Models like LLaVA-1.5, QwenVL-Chat, and Video-LLaVA by addressing the "inefficient attention" problem. Its use of FastV, a new method that optimizes these models by pruning visual tokens and learning adaptive attention patterns, leads to a significant reduction in computational costs without losing performance on tasks related to images and videos.
Researchers have created a generative AI worm called Morris II that can attack AI systems like ChatGPT, spreading autonomously while potentially stealing data. The worm uses “adversarial self-replicating prompts” to perpetuate and compromise AI email assistants, highlighting new cyberattack risks within the AI ecosystem. Security experts urge AI developers to take potential AI-driven threats seriously as AI applications become more autonomous and interconnected.
38 AI experts and 39 English experts were asked to rate and guess whether poems were written by an AI or a human. The human came in 1st place, while Bard, ChatGPT-4, and Claude came in 2nd, 3rd, and 4th places respectively, both in writing quality and their ability to fool respondents into believing their poems were authored by a human. English experts were far better at discerning which poems were written by AI, which points to a need for them to play a greater role in helping shape future versions of AI technology.
In a discussion about the need for AI regulation and transparent development practices with tech companies, Former President Barack Obama highlighted AI's potential risks and rewards and urged tech experts to engage in government roles to help shape thoughtful AI policies. The conversation also tackled First Amendment challenges and the necessity of a multi-faceted, adaptive regulatory approach for AI.
Perplexity has integrated Yelp's data to enhance chatbot responses with maps, reviews, and detailed information on restaurants to provide users with direct and rich local search results.