Search the archive

FACTUAL ACCURACY IN AI

Integrative Decoding is a method that enhances the factual accuracy of large language models, especially for open-ended tasks.

LIQUID FOUNDATION MODELS: OUR FIRST SERIES OF GENERATIVE AI MODELS

Liquid AI has announced its first generation of Liquid Foundation Models (LFMs), which deliver state-of-the-art performance with reduced memory usage. The LFMs, optimized for various hardware, include a 1B, 3B, and 40B model and are already available on platforms like LIQUID PLAYGROUND and soon on Cerebras. These models excel in handling sequential data and offer innovations in efficiency and scaling across industries such as financial services and biotechnology.

INTRODUCING COPILOT LABS AND COPILOT VISION

Microsoft is introducing Copilot Labs to trial advanced AI tools like Think Deeper and Copilot Vision.

OPENAI'S DEVDAY BRINGS REALTIME API AND OTHER TREATS FOR AI APP DEVELOPERS

OpenAI announced new tools at its 2024 DevDay, including a beta Realtime API for near real-time, low-latency, AI-generated voice responses.

MICROSOFT BRINGS AI-POWERED OVERVIEWS TO BING

Microsoft has launched Bing generative search, an AI-powered search feature that aggregates and summarizes web information.

👉 Watch now

AI ENHANCES SIMULATIONS WITH SMARTER SAMPLING

MIT CSAIL researchers have developed an AI-powered method for low-discrepancy sampling that improves the accuracy of simulations by distributing data points more uniformly.

ANNOUNCING FLUX1.1 [PRO] AND THE BFL API

FLUX1.1 [pro] has been released. It offers six times faster generation than its predecessor with improved image quality and performance. The new beta BFL API, which features advanced customization and competitive pricing, allows developers to integrate FLUX's capabilities. FLUX1.1 [pro] will be available on several platforms, enhancing scalability and efficiency.

OPENAI LAUNCHES NEW 'CANVAS' CHATGPT INTERFACE TAILORED TO WRITING AND CODING PROJECTS

OpenAI has launched "canvas," a new ChatGPT interface for writing and coding projects in a separate workspace. This move aligns with industry trends towards editable workspaces, competing with Anthropic's Artifacts and Cursor. canvas is initially available only to select users.

DISTRIBUTED TRAINING OF DEEP LEARNING MODELS

A great introduction to the challenges and some of the algorithms behind distributed training of many modern deep learning models, this post discusses the challenges and bottlenecks of training models that are too large for a single GPU.

COMFYGEN TEXT-TO-IMAGE

Instead of generating an image directly from a prompt, these authors generated a comfy UI node based workflow to generate an image. They found that this substantially improved the final output quality.

SKETCH-BASED IMAGE GENERATION

KnobGen is a new framework designed to make sketch-based image generation more accessible for users with different skill levels.

TINY IMAGE MODEL

AI researcher Ross Wightman has released a set of models trained on Imagenet-1k that are extremely small (sub 1 million parameters) but still perform somewhat well and are easy to tune.

ENTROPY BASED SAMPLING

Entropy based sampling and parallel Chain of Thought decoding are potential paths forward to match performance of reasoning models like o1.

A LIBRARY FOR GENERATIVE SOCIAL SIMULATION

Deepmind's Concordia repository can be used to simulate social interactions between individuals and groups at reasonable scale.

OPENAI'S ORIGINAL SIN

Recent departures of senior executives at OpenAI highlight ongoing internal conflicts between prioritizing product development and revenue versus research and safety. These issues stem from the company's foundational promise to develop AGI for the benefit of all humanity, which has proven challenging amidst financial and strategic pressures. OpenAI is negotiating a $6.5 billion funding round with conditions that could transform it into a for-profit entity.

THE ART OF THE OPENAI DEAL

OpenAI's revenue surged to $300M in August. The company is projecting $3.7B in annual sales for this year and $11.6B for next year, but it faces a $5B annual loss. Its rapid growth was fueled by the popularity of ChatGPT, which accounts for most of its revenue. Despite the growth, OpenAI is seeking more investors due to high operational costs and to evolve into a profitable entity.

WHAT COMES AFTER?

Californian Governor Gavin Newsom has vetoed the state's SB 1047 bill, which aimed to regulate large AI models. He emphasized the need for evidence-based regulations and warned against overly stringent rules that could stifle innovation. Newsom plans to develop empirical, science-based guidelines with experts like Dr. Fei-Fei Li.

ANTHROPIC HIRES OPENAI CO-FOUNDER DURK KINGMA

Durk Kingma, a co-founder of OpenAI, has joined Anthropic and will work remotely from the Netherlands.

OPENAI UNVEILS EASY VOICE ASSISTANT CREATION AT 2024 DEVELOPER EVENT

OpenAI announced major API updates at DevDay, including the Realtime API for speech-to-speech conversations and improvements to the Chat Completions API.

SORRY, GENAI IS NOT GOING TO 10X COMPUTER PROGRAMMING

Recent studies show that generative AI is not achieving the promised 10x improvement in coding productivity.

See what you can build with email data →

OPENAI'S LATEST FUNDING

OpenAI has raised $6.6B in new funding at a $157B post-money valuation to continue scaling AI. Investors include Microsoft and Nvidia.

GOOGLE ADDS A MULTI-FUNCTIONAL QUICK INSERT KEY AND NEW AI FEATURES TO CHROMEBOOK PLUS

Google has introduced new Chromebook models from Samsung and Lenovo featuring AI-powered tools. Key updates include a "Help me read" tool for summarizing PDFs and web content and live translation for video calls. Enhanced AI features are also coming to existing Chromebooks, such as AI writing tools and a built-in speech-to-text recorder.

BRAIN-LIKE COMPUTERS TACKLE THE EXTREME EDGE

BrainChip has announced the Akida Pico, a milliWatt-level neuromorphic chip designed for AI inference in power-constrained devices like wearables. The chip mimics brain activity to minimize energy use, targeting real-time processing needs in small user devices. Akida Pico joins similar edge-oriented neuromorphic chips from companies like Innatera and SynSense.

SCALING OPTIMAL LR ACROSS TOKEN HORIZON

This paper explores how to scale learning rate when you increase the amount of training data a model sees. Llama did exponential -0.28 and this paper suggests exponential -0.33.

EMBEDDING KNOWLEDGE GRAPHS

This paper introduces a new way to approach knowledge graph embedding by using group theory to incorporate uncertainty.

PERSON RE-ID WITH TEXTUAL PROMPTS

This paper presents a new method, Prompt-guided Feature Disentangling (ProFD), for addressing occlusion issues in person Re-Identification (ReID) tasks.

training and deploying models. Sign up and get started today!

LOCAL FILE ORGANIZATION

This tool uses Llama 3.2 3B and Llava-1.6 to organize your files on your computer into logical sections based on content.

NEW IMAGE RESTORATION ALGORITHM

Posterior-Mean Rectified Flow (PMRF) is an innovative algorithm for photo-realistic image restoration.

IMPROVE LLM COLLABORATION

RouterDC is a novel method that enhances collaboration between multiple large language models (LLMs) by using query-based routing. It employs contrastive learning to select the best model for each query, improving performance over existing routing methods.

AI CAN BEST GOOGLE'S BOT DETECTION SYSTEM, SWISS RESEARCHERS FIND

Researchers from ETH Zurich used AI to achieve a 100% success rate in solving Google's reCAPTCHA v2. This breakthrough highlights AI's capability to overcome image-based CAPTCHA systems. As bots become smarter, CAPTCHA's future as a security technology remains uncertain.

NEGLIGENCE LIABILITY FOR AI DEVELOPERS

This article discusses the need for a negligence-based approach to AI accountability, focusing on the human factors behind AI systems. It critiques current regulatory frameworks that overlook the role of AI developers and suggests California's AI safety bill as a potential model. The article also explores the challenges in defining reasonable care in AI and the implications of treating AI developers as professionals.

I AM TIRED OF AI

This author expresses frustration with the overuse and marketing of AI, particularly in software testing and conference proposals. They argue AI tools often prioritize speed over quality and lack the unique insight of human-generated work. Despite recognizing some beneficial uses, they criticize the proliferation of mediocre AI-generated content.

the blog

OPENAI TRAINING DATA TO BE INSPECTED IN AUTHORS' COPYRIGHT CASES

OpenAI has agreed to allow authors to review its training data amid copyright litigation.

THE FOUR SHORT TERM WINNERS OF AI

The global AI arms race is dominated by Big Tech firms, chipmakers like NVIDIA, intellectual property lawyers, and The Big 4 consulting firms.

BYTEDANCE WILL REPORTEDLY USE HUAWEI CHIPS TO TRAIN A NEW AI MODEL

ByteDance plans to train a new AI model using Huawei's Ascend 910B chips.

OPENAI DEV DAY 2024

There were lots of great announcements at OpenAI's Dev Day 2024, like vision model fine-tuning, real time API, prompt caching, and model distillation.

PIKA 1.5

Pika has released version 1.5 with more realistic movement, big screen shots, and Pikaffects.

GOV. NEWSOM VETOES CALIFORNIA'S CONTROVERSIAL AI BILL, SB 1047

Governor Gavin Newsom has vetoed SB 1047, a bill aimed at regulating AI development and imposing safety protocols on high-cost models. Newsom cited concerns that the bill's broad applicability to all large, computation-heavy models was not the best approach for effective AI regulation. Despite the veto, Newsom remains committed to developing AI safeguards, having recently signed multiple related bills and consulting experts for guidance.

MM1.5 FROM APPLE

Apple continues building its in-house capabilities with a strong multimodal foundation model.