News/AI Models
Meshy Just Released a New AI Model to Bring Sci-Fi Level AI to 3D Design
AI-powered 3D modeling takes a leap forward: Meshy, a startup in the AI design space, has released Meshy-4, its latest AI-powered 3D modeling tool, offering improved mesh geometry and a redesigned workflow that aims to revolutionize how designers and developers create virtual environments. Key improvements in Meshy-4: The new version addresses common issues with AI-generated 3D models by producing cleaner surfaces and enhanced geometric details, bringing the output closer to professional standards. The text-to-3D model generation process is now separatearticled into two stages: modeling and texturing, giving users more control over their final product. A new "Retry" option for the...
read Aug 26, 2024Why AI’s Insatiable Energy Demand Is More Nuanced Than You Think
AI energy consumption in context: The growing adoption of artificial intelligence has raised concerns about its energy use, but a closer examination reveals a more nuanced picture of AI's environmental impact and efficiency. Large language models and image generation AI systems consume significant energy during training, but their operational energy use (inference) is much lower. While data centers are major energy consumers, AI currently accounts for a small fraction of their total energy use. When considering the utility provided, AI may be more energy-efficient than alternative methods for many tasks. Comparing AI to traditional methods: In various applications, AI demonstrates...
read Aug 26, 2024Claude’s API Now Supports CORS Requests
Anthropic enables CORS support for Claude API: Anthropic has introduced CORS support for their JSON APIs, allowing developers to make direct calls to Claude language models from client-side applications. The new feature was quietly introduced through a pull request in the Anthropic TypeScript SDK, enabling browser-based usage of the API. Developers can now add the HTTP request header "anthropic-dangerous-direct-browser-access: true" to enable CORS support for the Anthropic API. Significance and potential use cases: This update opens up new possibilities for browser-based applications while also presenting potential security considerations. The feature is particularly useful for internal tools exposed to trusted users...
read Aug 26, 2024Incestual AI: How AI Models Training on Their Own Outputs May Lead to Disaster
AI-generated content poses unprecedented challenge: The proliferation of AI-generated content is creating a significant hurdle for AI companies, as they risk training new models on their own output, potentially leading to a deterioration in quality and diversity. OpenAI alone is estimated to produce about 100 billion words per day, contributing to the growing pool of AI-generated content on the internet. This surge in AI-created material raises concerns about the unintentional feedback loop that could occur when AI systems inadvertently ingest their own output during training. Researchers have identified a phenomenon called "model collapse," where the quality and diversity of AI-generated...
read Aug 25, 2024Frontier AI Models Could Cost $250B by 2027, Experts Predict
Scaling AI: The path to colossal models: Recent research and analysis suggest that by 2027, we could see the emergence of a $100 billion AI model, with further scaling beyond this point becoming less certain. Epoch AI's research forecasts AI training compute to reach 2e29 floating-point operations per second by 2030, requiring hardware investments of approximately $250 billion. This projected scale dwarfs current investments, being over five times Microsoft's annual capital expenditure. The study indicates no insurmountable technical barriers to this level of scaling, although there is high uncertainty surrounding various factors. Infrastructure challenges: Power availability and chip production present...
read Aug 25, 2024xAI Doubles Grok-2 Speed with Innovative Code Rewrite
Grok-2 and Grok-2 mini receive significant speed boost: xAI, Elon Musk's artificial intelligence company, has dramatically improved the performance of its large language models through a complete rewrite of the inference code stack. Two xAI developers, Lianmin Zheng and Saeed Maleki, rewrote the inference stack in just three days using SGLang, an open-source system for executing complex language model programs. The update resulted in Grok-2 mini becoming twice as fast as its previous version, while also enabling the larger Grok-2 model to run at a reasonable speed across multiple hosts. Both models experienced slight improvements in accuracy alongside their speed...
read Aug 25, 2024The ‘Plasticity Problem’ and LLM’s Inability to Learn Continuously
AI models face significant learning limitations: Recent research reveals that deep learning AI models, including large language models, struggle to incorporate new information without complete retraining. A study published in Nature by scientists from the University of Alberta highlights a major flaw in AI models' ability to learn continuously. Deep learning AI models, which find patterns in vast amounts of data, fail to function effectively in "continual learning settings" where new concepts are introduced to existing training. Attempting to teach an existing deep learning model new information often requires retraining it entirely from scratch. The plasticity problem: When new information...
read Aug 25, 2024Tech Giants Push for Open-Source AI to Fuel Innovation
AI industry leaders advocate for open-source models: Mark Zuckerberg and Daniel Ek make a compelling case for open-sourcing AI software, particularly in Europe, to prevent power concentration and foster innovation. Zuckerberg and Ek argue that open-sourcing AI models creates a level playing field and ensures power isn't concentrated among a few large players. The approach aligns with Meta's recent shift in priorities, focusing more on AI investments rather than the "metaverse." This stance marks a notable change in perception for Zuckerberg, who has faced criticism for past decisions but is now gaining support for his AI-focused strategy. The future of...
read Aug 25, 2024How Google’s 3 New AI Models Stack Up Against Each Other
Google's Gemini AI: A new frontier in language models: Google's latest large language model, Gemini, comes in three distinct versions - Ultra, Pro, and Nano - each tailored for different use cases and computational environments. Gemini Nano: AI in your pocket: This lightweight version is designed to run directly on mobile devices, offering on-device AI capabilities without compromising user privacy or requiring constant internet connectivity. Gemini Nano comes in two variants: Nano-1 with 1.8 billion parameters and Nano-2 with 3.25 billion parameters. It powers on-device AI features such as call notes on Pixel phones, showcasing its ability to perform complex...
read Aug 25, 2024Abacus.AI Releases Open-Source AI Model ‘Dracarys’ for Coding
Generative AI for coding gets a boost: Abacus.ai has released Dracarys, a new family of open-source large language models (LLMs) specifically optimized for coding tasks. Dracarys applies a fine-tuning "recipe" to existing LLMs, significantly improving their coding abilities. The initial release focuses on the 70B parameter class of models, including Meta's Llama 3.1 and Qwen-2 72B. Abacus.ai has previously released Smaug-72B, a general-purpose LLM, showcasing their penchant for using dragon-inspired names from popular fiction. Performance improvements: Benchmarks indicate substantial enhancements in coding capabilities for models treated with the Dracarys recipe. LiveBench coding scores show an increase from 32.67 to 35.23...
read Aug 25, 2024Microsoft’s Small Phi-3.5 Model Outperforms Gemini and GPT-4o in STEM
Microsoft unveils Phi-3.5 small language model: Microsoft has released its latest iteration of small language models, Phi-3.5, available in three sizes and free to download. Model specifications and performance: Phi-3.5 comes in 3.8 billion, 4.15 billion, and 41.9 billion parameter versions, with the smallest model trained on 3.4 trillion tokens of data using 512 Nvidia H100 GPUs over 10 days. The model excels in reasoning tasks, second only to GPT-4o-mini among leading small models. Phi-3.5 significantly outperforms Llama and Gemini on math benchmarks. A vision-capable version of the model can process and understand images. The largest version utilizes a mixture...
read Aug 25, 2024How Lab-Grown Mini Brains Could Power AI of the Future
Breakthrough in biocomputing: Scientists are exploring the potential of brain organoids, tiny lab-grown neural structures, to power AI systems with unprecedented efficiency and sustainability. The AI energy crisis: Current artificial intelligence systems consume enormous amounts of energy, raising concerns about their long-term sustainability. • OpenAI's ChatGPT alone requires 500,000 kilowatts of power daily to process 200 million user prompts. • The scale of resources needed to fuel the global AI boom is becoming increasingly unsustainable. Biological solutions to technological problems: Researchers are turning to neuroscience and biotechnology to address the energy demands of AI systems. • Companies like Cortical Labs...
read Aug 25, 2024How LLMs and the Human Brain Process Information Differently
The fundamental question: Can large language models (LLMs) truly think and reason like humans, or are their capabilities fundamentally different from human cognition? This question has sparked intense debate in the fields of artificial intelligence, cognitive science, and neuroscience as researchers seek to understand the similarities and differences between AI and human intelligence. The comparison between LLMs and human cognition centers on the critical cognitive function of inference, which allows for abstract reasoning and the application of knowledge across diverse contexts. The role of the hippocampus: The human brain's hippocampus plays a crucial role in enabling abstract reasoning and inference...
read Aug 23, 2024How Piramidal is Using AI to Decode Human Brain Waves
Pioneering AI for brain wave interpretation: Piramidal, a Y Combinator-backed startup, is developing a groundbreaking foundation model to decode complex brain language and interpret EEG data, with potential applications across various medical fields and beyond. The company's AI model is being trained on brainwave data in a manner similar to how ChatGPT is trained on text, making it the largest model ever trained on EEG data. Piramidal recently secured $6 million in funding from Y Combinator, Adverb Ventures, Lionheart Ventures, and angel investors including founders of Intercom, Plangrid, and Guilded. Addressing challenges in EEG interpretation: The startup aims to overcome...
read Aug 23, 2024New Research Suggests AI Models Can’t Learn as They Go Along
AI models face limitations in continuous learning: Recent research reveals that current artificial intelligence systems, including large language models like ChatGPT, are unable to update and learn from new data after their initial training phase. A study by researchers at the University of Alberta in Canada has uncovered an inherent problem in the design of AI models that prevents them from learning continuously. This limitation forces tech companies to spend billions of dollars training new models from scratch when new data becomes available. The inability to incorporate new knowledge after initial training has been a long-standing concern in the AI...
read Aug 23, 2024Ideogram 2.0 Raises the Bar for AI-Generated Text in Images
Advancing AI image generation: Ideogram's release of version 2.0 of its text-to-image AI model marks a significant leap forward in the field, challenging established players like Midjourney and DALL-E 3. The new model boasts enhanced text rendering capabilities, addressing a longstanding challenge in AI-generated imagery and opening up new possibilities for rapid prototyping in advertising and branding. A customizable color palette feature has been introduced, allowing for greater brand consistency in generated images. Ideogram has launched a public beta API, enabling developers to integrate this cutting-edge technology into their own applications and workflows. The company has adopted a competitive pricing...
read Aug 23, 2024Midjourney Offers 25 Free AI Images as Website Opens to All
Midjourney, a leading AI image generation platform, has taken a significant step towards wider accessibility by opening its website to all users, offering 25 free AI-generated images to newcomers. Expanded access and free offerings: Midjourney's decision to open its website to all users marks a strategic shift in its approach to user acquisition and engagement. Previously, the website was limited to an "alpha" version, accessible only to users who had generated a certain number of images. New users can now sign up using either a Google account or a Discord account, simplifying the onboarding process. The platform is offering approximately...
read Aug 23, 2024NVIDIA’s New AI Model Makes Conversations with Game Characters More Real
NVIDIA unveils groundbreaking on-device AI model for gaming: NVIDIA has introduced Nemotron-4 4B Instruct, a small language model designed to enhance character interactions in video games, marking a significant advancement in digital human technologies. The new model is part of NVIDIA's ACE (Avatar Cloud Engine) suite, which aims to improve speech, intelligence, and animation in gaming through generative AI. Nemotron-4 4B Instruct is optimized to run directly on GeForce RTX AI PCs and laptops, offering faster response times and reduced memory usage. This development allows game developers to leverage the power of over 100 million GeForce RTX-powered devices to create...
read Aug 23, 2024AI Fact-Checking and LLMs’ Role in the Misinformation Battle
The rise of large language models: The emergence of advanced AI tools like ChatGPT and Google's Gemini has revolutionized natural language generation, offering both immense potential and significant challenges in terms of factual accuracy. Large language models (LLMs) have demonstrated remarkable capabilities in generating human-like text, making them valuable for various applications. However, these models are prone to producing false, erroneous, or misleading content, a phenomenon known as "hallucinations." The ability of LLMs to generate convincing yet false content at scale poses a substantial societal challenge, potentially deceiving users and spreading misinformation. Factuality challenges and implications: The tendency of LLMs...
read Aug 23, 2024GPT-4 Matches Radiology Residents in Musculoskeletal Imaging Accuracy
Comparing ChatGPT and radiologists in musculoskeletal imaging: A recent study led by researchers from Osaka Metropolitan University's Graduate School of Medicine evaluated the diagnostic accuracy of ChatGPT against radiologists in musculoskeletal imaging cases. The study, conducted by Dr. Daisuke Horiuchi and Associate Professor Daiju Ueda, aimed to assess the potential of generative AI models like ChatGPT as diagnostic tools in radiology. Researchers analyzed 106 musculoskeletal radiology cases, including patient medical histories, images, and imaging findings. Two versions of the AI model, GPT-4 and GPT-4 with vision (GPT-4V), were used to generate diagnoses based on the case information. The same cases...
read Aug 23, 2024The Open Source Initiative Creates New Definition for Open-Source
Defining open-source AI: The Open Source Initiative (OSI) has unveiled a new definition for open-source AI systems, aiming to provide clarity in a field where the concept was previously ambiguous. The definition outlines key criteria for AI systems to be considered open-source, including unrestricted use, inspectability, modifiability, and shareability. Transparency requirements extend to training data, source code, and model weights, ensuring a comprehensive understanding of the AI system's components. The definition stipulates that sufficient information must be provided to allow a skilled person to recreate a substantially equivalent system using the same or similar data. Collaborative effort and development process:...
read Aug 21, 2024Zed Unveils AI-Powered Coding Assistant with Claude 3.5 Integration
Introducing Zed AI: Zed, a company known for its expertise in text manipulation and editor development, has launched Zed AI, a hosted service that integrates AI-powered coding assistance into their text editor. Key features and collaborations: Zed AI is powered by Anthropic's Claude 3.5 Sonnet language model, offering AI-enabled coding support within the Zed editor. The collaboration with Anthropic arose from mutual interest, with Anthropic engineers recognizing the value of Zed's text-centric interface. Zed has implemented Anthropic's Prompt Caching beta, resulting in faster responses and reduced costs, even with large amounts of code in the context window. The service is...
read Aug 21, 2024Microsoft Unveils 3 Powerful New Open-Source AI Models
Breakthrough in AI development: Microsoft has unveiled three new powerful AI models in its Phi series, marking a significant advancement in the field of artificial intelligence and machine learning. The new models, Phi-3.5-mini-instruct, Phi-3.5-MoE-instruct, and Phi-3.5-vision-instruct, are designed for various tasks ranging from basic reasoning to complex vision-related problems. These models are now available on Hugging Face under an MIT License, allowing for both commercial use and modification, which could potentially accelerate AI innovation across various sectors. In benchmark tests, the Phi-3.5 models have demonstrated impressive performance, surpassing some models from tech giants like Google, Meta, and OpenAI in certain...
read Aug 21, 2024Nvidia Creates Mini Version of Llama 3.1 Model That Punches Far Above Its Weight
Nvidia's breakthrough in small language models: Nvidia researchers have developed Llama-3.1-Minitron 4B, a compressed version of the Llama 3 model that rivals larger models while being more efficient to train and deploy. The new model leverages recent advances in pruning and distillation techniques to create a powerful small language model (SLM) for resource-constrained devices. Llama-3.1-Minitron 4B's performance is comparable to larger models and equally sized SLMs, despite being trained on a significantly smaller dataset. Key techniques: Pruning and distillation: These methods are crucial for creating smaller, more efficient language models without sacrificing performance. Pruning involves removing less important components of...
read