Search the archive

OPENAI STOPPED STATE-SPONSORED ACTORS

OpenAI discovered and terminated accounts affiliated with nation-states using GPT models for malicious cases.

ANDREJ KARPATHY LEAVES OPENAI

Andrej Karpathy, a founding member of OpenAI and former leader of Tesla's Autopilot team, announced his departure from OpenAI to focus on personal projects. His exit is not due to any specific event, issue, or drama.

NVIDIA’S NEW CHATBOT RUNS LOCALLY ON YOUR PC

NVIDIA has launched a demo of Chat with RTX, a local PC-run chatbot that can analyze and summarize personal data from various file formats and even integrate knowledge from YouTube videos, with the aim of creating a more personalized digital assistant experience. This innovation, which does not require cloud processing or an internet connection for sensitive data, is currently limited to Windows PCs with specific NVIDIA GPUs.

VIDEO LANGUAGE MODEL THAT CAN ANSWER QUESTIONS ABOUT HOUR LONG VIDEOS

This model uses ring attention and a fine-tuned 7B parameter model to accurately answer million token video length questions. It outperforms commercial VLMs and has extremely accurate performance on retrieval benchmarks.

LUMIERE TEXT TO VIDEO MODEL

Google has a new text-to-video model that can help take images and styles as input. It uses a novel “space-time UNet” to diffuse everything at once.

CRAFTING LONG VIDEOS WITH CREATIVE AI TRANSITIONS

SEINE is an innovative video diffusion model guided by textual descriptions designed to extend short AI-generated video clips into longer, story-level sequences with seamless and imaginative scene transitions.

Write better code, faster. Book a demo or try Cody yourself.

MAGNET AUDIO GENERATION MODEL

Facebook released a state-of-the-art open-source audio model that is 7x faster than alternatives without sacrificing quality. It can generate music and sound effects. The paper is now available.

PRIVACY IN LARGE LANGUAGE MODELS

Researchers conducted a comprehensive study on whether it's possible to tell if a data point was used to train a large language model. They tested various models and found that, for the most part, these attacks are no better than random guesses.

TRUSTWORTHY AUTONOMOUS AGENTS

This project introduces advanced methods to ensure that autonomous intelligent agents, which are crucial in life-dependent applications, remain trustworthy and ethical even as they evolve.

WHY "CHAT OVER YOUR DATA" IS HARDER THAN YOU THINK

Building domain-specific, chat-based LLM applications and copilots is harder than people think. Some of the challenges include achieving robust performance, handling complex data and complex queries, and enabling robust data retrieval for LLM-based chat applications.

TEXT-DRIVEN IMAGE EDITING WITHOUT MASKS

Researchers have developed a unique method for editing images based on text prompts. This approach uses a text-to-image model and a bounding box generator to identify regions for editing, allowing for changes without needing masks or sketches.

VIDEO ANNOTATOR FOR ENHANCED VIDEO CLASSIFICATION

The Video Annotator framework involves domain experts directly in the annotation process. This new approach, integrating human expertise with zero-shot and active learning techniques, improves model accuracy and efficiency.

the Onehouse + Confluent webinar

PRESSPULSE (PRODUCT)

Build press mentions using AI.

NVIDIA IS NOW WORTH AS MUCH AS THE ENTIRE CHINESE STOCK MARKET

Thanks to the AI boom, NVIDIA's market cap has reached $1.7 trillion, equivalent to the combined market cap of all Chinese firms listed on the Hong Kong Stock Exchange.

REOR

An AI note-taking app that runs models locally.

YOUR REFERRAL LINK - https://tldr.tech/ai?ref=2613930

STABILITY AND WURSTCHEN RELEASE NEW TEXT TO IMAGE MODEL

The Stable Cascade model is a new research-only text-to-image model that outperforms SDXL in aesthetics, prompt faithfulness, and text quality. It was trained using the Würstchen architecture and has a multi-step architecture that makes tuning easy.

OPENAI TESTING MEMORY IN CHATGPT

OpenAI is testing a new feature that allows ChatGPT to remember facts across conversations. This can be switched off if desired. It will allow for a higher measure of personalization when interacting with the chat system.

LETTER FROM THE YOUTUBE CEO: 4 BIG BETS FOR 2024

YouTube is investing in diverse revenue streams for creators. The platform witnessed a 50% increase in the use of channel memberships. It is fostering creator support networks through initiatives like the Creator Collective. Efforts are underway to help policymakers understand and recognize the economic and entertainment value of creators.

BOOSTING MATH SKILLS IN AI

Researchers have developed a new technique to enhance language models' ability in math by using base models to select high-quality mathematical content autonomously.

A NOVEL IMAGE SEGMENTATION

Researchers have introduced a new technique for image segmentation that identifies specific parts of an image, like a dog, using only basic image labels. They tackled the challenge of a network recognizing multiple instances of the same object by introducing a novel method that refines the process and reduces errors.

SAFEGUARDING PRIVACY WITH LARGE LANGUAGE MODELS

CUDA FOR AMD

ZLUDA is a port of CUDA for AMD chips. It is in alpha but has been shown to run many graphics software packages like Blender successfully.

ENHANCING TRANSLATION ACCURACY

GenTranslate is a new approach that uses large language models to improve the quality of translations. It focuses on the best possible translations generated by foundational models. The method has proven to outperform current leading translation models in tests.

DESIGN2CODE

Design2Code is an open-source project that converts various web design formats, including sketches, wireframes, Figma, and XD, into clean and responsive HTML/CSS/JS code.

AN INTERVIEW WITH META’S AI CHIEF YANN LECUN

Yann LeCun discusses AGI, open-source, and AI risk with Time Magazine.

SAM ALTMAN SEEKING TRILLIONS FOR AI CHIP FABRICATION FROM UAE, OTHERS

OpenAI CEO Sam Altman is in talks with investors to raise up to $7 trillion for AI chip manufacturing. The project aims to address GPU scarcity as the chips are crucial for training and running large language models. Potential investors include sovereign wealth funds and government entities, which may raise geopolitical concerns. The US has been cautious about allowing foreign control over the supply of microchips - it recently announced a $5 billion investment in R&D to advance US-made semiconductor technologies.

META TO DEPLOY IN-HOUSE CUSTOM CHIPS THIS YEAR TO POWER AI DRIVE

Meta is planning to deploy a new version of its custom AI chip in data centers this year, aiming to reduce reliance on Nvidia chips and control costs for running AI workloads.

and networking. Save your spot in 10 seconds

GOOGLE'S €25M AI SKILLS TRAINING INITIATIVE IN EUROPE

Google launched a skills training plan for Europe. It offers free courses and startup opportunity credits.

SGLANG

A structured generation language designed for LLMs.

GALILEO AI (PRODUCT)

Generate interface designs at lightning speeds.

YOUR REFERRAL LINK - https://tldr.tech/ai?ref=2613930

REKA HAS A NEW POWERFUL MULTI-MODAL MODEL

AI startup Reka has announced a 21B parameter model that was trained completely from scratch. It is natively multimodal and surpasses Llama 70B and Mixtral on a number of key benchmarks. Importantly, given where the founders of the company used to work, it also matches Gemini Pro. The model is available for use in Reka's playground.

APPLE RELEASES MGIE, A REVOLUTIONARY AI MODEL FOR INSTRUCTION-BASED IMAGE EDITING

Apple has launched MGIE, an open-source AI model for image editing that uses multimodal large language models to interpret text instructions and execute pixel-level edits.

DEEPMIND FRAMEWORK OFFERS BREAKTHROUGH IN LLMS’ REASONING

DeepMind and USC researchers have developed SELF-DISCOVER, a new framework that significantly improves LLM reasoning capabilities, outperforming traditional methods by up to 32% on tasks with models like GPT-4. The two-stage framework allows LLMs to form and utilize reasoning structures autonomously. This advancement represents a major step toward more sophisticated AI problem-solving abilities.

CONTROLLING REAL WORLD OBJECTS WITH FLUID IN RL

"Box o Flows" introduces a novel experimental system for testing reinforcement learning (RL) algorithms in dynamic real-world settings, addressing the challenge of simulating complex fluid dynamics for RL applications. It showcases the ability of model-free RL algorithms to generate complex behaviors from simple rewards and enhances data efficiency via offline RL and paves the way for broader RL applicability in complex systems.

WEBLINX CONVERSATIONAL WEB SCRAPING DATASET

Weblinx is a dataset of 100k web-based interactions in a conversational format. It was released to improve the study of language model guided web-based navigation.

3D AVATARS WITH IMPLICITDEEPFAKE1

This paper introduces ImplicitDeepfake1, a new approach that combines deepfake technology with Neural Radiance Fields (NeRFs) and Gaussian Splatting (GS) to create highly realistic 3D avatars.

during the breaks. Sign up or miss out (it’s 100% free)

GUIDING INSTRUCTION-BASED IMAGE EDITING VIA MULTIMODAL LARGE LANGUAGE MODELS

Apple released code that uses multimodal language models to improve human-provided natural language edits to images.

LAG-LLAMA: TOWARDS FOUNDATION MODELS FOR PROBABILISTIC TIME SERIES FORECASTING

Lag-Llama introduces a breakthrough in time series forecasting with a transformer-based model that excels in zero-shot and few-shot generalization, pre-trained on a broad spectrum of time series data. It surpasses existing deep learning methods in state-of-the-art performance through fine-tuning on minimal data, marking a significant advancement in foundation models for time series analysis.

LEARNING TO FLY

A repository for training end-to-end control policies using deep reinforcement learning. The training is done in simulation and is finished within seconds on a consumer-grade laptop. The trained policies generalize and can be deployed on real quadrotors.

MASSED MUDDLER INTELLIGENCE

The concept of Massed Muddler Intelligence (MMI) represents a shift from traditional monolithic AI scaling towards a model based on distributed, agent-based systems that learn and adapt in real time. Grounded in principles of embodiment, boundary intelligence, temporality, and personhood, MMI advocates for AI development that emphasizes scalable, interactive agents with a degree of autonomy and mutual governance, moving away from the current focus on accumulating larger datasets and computational resources.