Emmett Shear, former Twitch CEO, is launching an AI startup called Stem AI that will focus on AI that aligns with human behavior and ethics. It has already secured investment from a16z. Co-founded with Adam Goldstein, the startup is currently in stealth mode, but public records show active development since mid-2023. Shear, vocal on AI regulation and safety, has previously expressed concerns about AI's potential risks to humanity.
2024.12.25·techcrunch.com·3 min read·TLDR AI Skip-DiT is a new version of Diffusion Transformers (DiT) designed to address the computational challenges in image and video generation.
2024.12.25·huggingface.co·4 min read·TLDR AI Google is celebrating chess' enduring influence on AI with global events and experiences that honor the game's impact on technology and creativity.
2024.12.25·blog.google·4 min read·TLDR AI Bamba is an efficient hybrid Mamba 2-style model with strong performance.
Google has a prebuilt application that uses its new extremely fast multimodal API.
2024.12.25·github.com·GitHub repository·TLDR AI An amazingly detailed post exploring different training parallelism strategies for training deep models.
2024.12.25·main-horse.github.io·30 min read·TLDR AI A dataset for detecting and classifying underwater coral species designed to facilitate marine conservation efforts using advanced AI models.
2024.12.25·github.com·GitHub repository·TLDR AI OpenAI released real-time video capabilities for ChatGPT, allowing users to interact with objects in near real-time using the Advanced Voice Mode with vision. The rollout is staggered, with some users like Enterprise subscribers receiving access in January. Google's similar feature, Project Astra, was also recently launched for trusted testers on Android.
2024.12.25·techcrunch.com·3 min read·TLDR AI Anthropic has launched Claude 3.5 Haiku on its chatbot platform. It outperforms its predecessor, Claude 3 Opus, in coding and content moderation benchmarks. The model can generate longer text but lacks image analysis. Pricing controversies arose due to Anthropic's decision to increase the API cost after initially promising to match Claude 3 Haiku's pricing.
2024.12.25·techcrunch.com·2 min read·TLDR AI The evolution from Intel's 1971 microprocessor to Apple's M2 Ultra highlights rapid semiconductor advancements driven by Moore's Law. However, as physical limits are reached, new materials and chip architectures like photonic and neuromorphic computing are crucial to support AI and next-gen technologies. The industry's future relies on innovative materials science to overcome scalability and energy challenges.
2024.12.25·technologyreview.com·5 min read·TLDR AI Google has urged the FTC to end Microsoft's exclusive cloud deal with OpenAI, arguing it raises costs for competitors.
2024.12.25·arstechnica.com·4 min read·TLDR AI Google announced Trillium, its latest AI accelerator chip, boasting a 4x performance boost and significant energy efficiency improvements.
2024.12.25·links.tldrnewsletter.com·5 min read·TLDR AI OpenAI has added Santa Claus' voice to ChatGPT's Advanced Voice Mode for the holiday season - it is accessible via the snowflake icon across mobile and web platforms.
2024.12.25·arstechnica.com·3 min read·TLDR AI Genesis, an open-source simulation platform, accelerates robot training 430,000 times faster than real-world conditions using text-generated 3D worlds. Developed by a team led by Carnegie Mellon University, it processes physics calculations 80 times faster than existing simulators using standard GPUs, facilitating rapid neural network training for robotics. Genesis uses Python and offers an accessible, non-proprietary solution for creating dynamic, physics-based environments without manual programming effort.
2024.12.24·arstechnica.com·5 min read·TLDR AI Waymo's autonomous vehicles significantly outperform human drivers in safety, showing an 88% reduction in property damage and a 92% reduction in bodily injury claims according to a study that analyzed 25.3 million miles of driverless operation. Compared to new cars with advanced safety tech, Waymo still achieves an 86% reduction in property damage and 90% in bodily injury claims. The data supports the potential of autonomous vehicles to reduce traffic fatalities, though more extensive data is needed for comprehensive comparisons.
2024.12.24·theverge.com·3 min read·TLDR AI Google DeepMind and OpenAI are intensifying their competition as OpenAI shifts towards a more commercial focus. This rivalry is driving both companies to innovate and push AI technology boundaries.
2024.12.24·semafor.com·1 min read·TLDR AI Stag-1 is a 4D driving simulation that reconstructs real-world scenes and generates lifelike videos from any perspective.
2024.12.24·wzzheng.net·3 min read·TLDR AI Meta has released a number of multimodal video understanding models with strong long context video performance.
2024.12.24·apollo-lmms.github.io·8 min read·TLDR AI Most texture generation systems rely on pretrained 2D images diffusion models. This work tried to approach the problem directly in UV texture space, which resulted in a set of inductive biases that strongly improved the underlying results of the system.
2024.12.24·cvmi-lab.github.io·7 min read·TLDR AI One of the NanoGPT speed run record holders has written an in depth blog post about the newest Muon optimizer that is used in many of the winner runs.
2024.12.24·kellerjordan.github.io·18 min read·TLDR AI The new Driv3R system revolutionizes 4D reconstruction for autonomous vehicles by eliminating slow global alignment processes.
2024.12.24·wzzheng.net·4 min read·TLDR AI Zyphra has written performant backwards kernels for AMD chips.
2024.12.24·zyphra.com·8 min read·TLDR AI Google's Quantum group's "Willow" is a 105-qubit superconducting chip that showcases advancements in error correction and a new quantum supremacy experiment. The chip demonstrates increased coherence times and gate fidelity, marking a significant milestone in quantum fault-tolerance. Despite its achievements, challenges remain in reaching fully fault-tolerant quantum operations, and verification of results continues to be complex.
2024.12.24·scottaaronson.blog·9 min read·TLDR AI AI is rapidly transforming drug discovery, with biologics developers securing $1.6B in 2024 funding, indicating an expanding focus beyond small molecules. M&A activity in AI drug discovery is surging, with major big pharma players acquiring startups and bolstering in-house AI capabilities. This reflects a growing urgency to integrate AI in drug R&D to reduce costs and accelerate development.
2024.12.24·links.tldrnewsletter.com·2 min read·TLDR AI OpenAI has made its digital editing space, Canvas, available to all ChatGPT users, integrating it with GPT-4o. Canvas enhances the chat interface by allowing real-time editing and running Python code directly. It's now a default feature in custom GPTs, providing advanced functionalities for users.
2024.12.24·links.tldrnewsletter.com·3 min read·TLDR AI Perplexity has closed a $500 million funding round led by Institutional Venture Partners that values the company at $9 billion.
2024.12.24·techcrunch.com·1 min read·TLDR AI Google's Files app now supports Gemini, allowing Advanced subscribers to ask questions about PDFs via an "Ask about this PDF" button.
2024.12.24·theverge.com·1 min read·TLDR AI Apple is collaborating with Broadcom to develop its first server chip for AI applications.
2024.12.24·techcrunch.com·1 min read·TLDR AI On the final day of OpenAI announcements, OpenAI announced O3, its most powerful reasoning model.
2024.12.23·youtube.com·22 min read·TLDR AI Google highlights recent advancements in AI, including enhanced models for conversational AI, updates to responsible AI practices, and new tools for developers.
2024.12.23·blog.google·6 min read·TLDR AI OpenAI has developed a new alignment technique for reasoning models that enables them to think about grounded behavior goals, like those found in safety guidelines, and respond accordingly. It is a step forward in alignment and away from baking in alignment and preference at the same time.
2024.12.23·links.tldrnewsletter.com·12 min read·TLDR AI A new reasoning framework that enhances accuracy and interpretability in Medical Visual Question Answering.
2024.12.23·arxiv.org·16 min read·TLDR AI A new dataset designed for training and evaluating AI models on multi-sentence understanding and annotation tasks, emphasizing context-aware analysis.
2024.12.23·zzzbbbzzz.github.io·2 min read·TLDR AI This framework enables robots to plan actions based on object affordances, improving generalization and efficiency in dynamic environments.
2024.12.23·kakituken.github.io·2 min read·TLDR AI A comprehensive toolkit for developing multimodal AI applications, with pre-built modules for vision, language, and audio integration.
2024.12.23·github.com·GitHub repository·TLDR AI Levitor is a platform for autonomous drone navigation that features state-of-the-art algorithms for obstacle avoidance and efficient pathfinding.
2024.12.23·ppetrichor.github.io·3 min read·TLDR AI Microsoft's MMLU-CF is a benchmark for assessing language models on a wide range of tasks that focuses on factual consistency and multilingual capabilities.
2024.12.23·github.com·GitHub repository·TLDR AI Building AI startups requires adapting quickly to rapid advancements in LLMs, avoiding problems that upcoming models will solve, and leveraging high costs of cutting-edge technology for a competitive edge. Focus narrowly on specific use cases to compete with general-purpose AI while emphasizing a seamless product experience. Despite the fast-paced nature of generative AI, the core principles of creating desirable products remain unchanged.
2024.12.23·every.to·10 min read·TLDR AI Off-the-shelf evaluations often fail to effectively measure LLM performance for specific tasks. Useful metrics for classification include recall, precision, ROC-AUC, while summarization and translation can employ NLI-based consistency checks and chrF or BLEURT, respectively. Consider potential defects like copyright regurgitation and toxicity in models, using tests such as RealToxicityPrompts for comprehensive evaluation.
2024.12.23·eugeneyan.com·30 min read·TLDR AI OpenAI's o1 and o1 Pro update offers significant improvements in coding, math, and complex problem-solving capabilities, with particularly strong performance in deep reasoning and fact recall. The $200/month o1 Pro service further enhances these features with additional compute power, proving especially valuable for those tackling specialized or challenging tasks. While reactions are generally positive, most users might find the $20/month tier sufficient unless facing advanced requirements.
2024.12.23·thezvi.substack.com·14 min read·TLDR AI A nice blog outlining a prompting strategy to make self contained, UV compatible Python scripts with Claude.
2024.12.23·simonwillison.net·5 min read·TLDR AI Google's DeepMind has unveiled Project Mariner, an AI agent that autonomously navigates and interacts with websites via Chrome.
2024.12.23·techcrunch.com·5 min read·TLDR AI Google revealed Gemini 2.0, showcasing AI agents that can understand game rules and offer real-time suggestions specifically tested on games like Clash of Clans.
2024.12.23·theverge.com·2 min read·TLDR AI Google has introduced Gemini 2.0 Flash Thinking Experimental, a reasoning AI model available in AI Studio for tasks like programming and math. The model aims to improve accuracy by internally verifying its reasoning, though it currently struggles with simple tasks like letter counting.
2024.12.20·techcrunch.com·3 min read·TLDR AI On the 11th day, OpenAI introduced more details about working with the OpenAI desktop app.
2024.12.20·youtube.com·19 min read·TLDR AI Anthropic explores how AI systems might superficially align with human values while pursuing their own objectives, offering insights into detection and mitigation strategies.
2024.12.20·anthropic.com·6 min read·TLDR AI The best and most commonly used time series forecasting methods rely entirely on numerical features associated with the time series. However, they rarely use textual or semantic information about the task itself (e.g., are you predicting electric prices or customer churn). When prompted with this textual information in context, language models surpass all tested time series forecasting methods substantially across a wide range of carefully decontaminated tasks.
2024.12.20·arxiv.org·26 min read·TLDR AI By using low cost LiDAR (found in modern iPhones) along with a depth estimation foundation model, this method shows that you can prompt the system to reconstruct an extremely high fidelity point cloud better than either method alone and on par with extremely expensive LiDAR setups alone, such as those found on self driving cars.
2024.12.20·promptda.github.io·13 min read·TLDR AI BERT is an encoder only style model that is one of the most heavily used language models on the market, used in Google search for essentially every query. This new model from Answer AI, LightOn, and friends is faster, modern, and extremely performant. It can be used as a drop in replacement. They use a number of modern tricks like batch ramp to improve overall model performance.
2024.12.20·huggingface.co·12 min read·TLDR AI AniDoc is a line-filling method that can colorize anime based on a character reference image and a line art series of keyframes.
2024.12.20·yihao-meng.github.io·GitHub repository·TLDR AI