The smart home tech that saves you money

RYSE is the smart-home startup that’s creating a brand new category: technologies that automate your existing window coverings.

What’s different about their products?

Existing smart shade tech requires you to replace your shades, but RYSE lets you transform your existing shades into automated ones.

It’s a simple, 5 minute installation that gives you access to remote shade controls, smart home integrations, and scheduling - all at a fraction of the cost of motorized shades.

Their products are currently on sale, and you can get them here.

Today:

• Meta Launched MovieGen

• ChatGPT Canvas for Edits

• Google Adds Ads to AI Summaries

• BFL Launches Faster Flux 1.1 Pro

• New Languages for Gemini Live Users

• Cohere Boosts AI Fine-Tuning Efficiency

Zuck Announces Meta's NEW AI Model... it's 🔥 up…

Meta’s new AI model, MovieGen, introduces groundbreaking video creation and editing features coming to Instagram next year. It allows users to generate videos from text, upload images for personalized videos, and edit videos using text commands for detailed adjustments like adding effects or changing backgrounds. 

While not the best on the market yet, Meta’s massive user base and technological resources give it a significant edge. Additional features include sound generation for videos, adding another layer of creativity. Meta's vast computing power and resources could make this AI tool a strong contender in the AI video space soon.

WATCH THE VIDEO ON YOUTUBE

ChatGPT Canvas for Edits

OpenAI has launched "ChatGPT Canvas," a new feature that allows users to directly edit and modify AI-generated text in a side-by-side view. This feature, built on the GPT-4o model, aims to streamline editing for writing and coding tasks by offering easy adjustments without generating new responses. 

Canvas also supports code review and bug fixes. It competes with Anthropic's Claude Artifacts, which offers a similar interface for text and code edits. Canvas is available to ChatGPT Plus and Teams users, with wider availability planned.

Google Adds Ads to AI Summaries

Google has added ads to its AI-generated search summaries. When users ask questions with a commercial angle, products related to the search will appear under a "sponsored" label. This feature is currently rolling out in the U.S. on mobile and aims to help users quickly find relevant products. 

Google has been testing the feature since May and found it beneficial for connecting users with businesses. Additionally, Google is adjusting how it displays sources in AI summaries and testing AI-organized search pages, which are available for recipe-related searches in the U.S. on mobile.

BFL Launches Faster Flux 1.1 Pro

Black Forest Labs (BFL) has launched Flux 1.1 Pro, a faster and improved version of its text-to-image AI model. It features six times the speed of its predecessor, with better image quality and prompt adherence. Alongside this, BFL released an API, allowing developers to integrate Flux into their apps for a fee. The model is accessible through third-party platforms and has topped benchmarks for image quality. 

BFL's API offers scalable, customizable tools for various industries. The company, backed by major investors, is also planning to expand into AI-driven text-to-video systems.

New Languages for Gemini Live Users

Google’s Gemini Live, initially launched in English, is expanding to include French, German, Portuguese, Hindi, and Spanish. More languages will follow in the coming weeks, with over 40 expected. Initially exclusive to Pixel 9 users, Gemini Live is now free for all Android users. 

In addition to voice chat, Gemini will soon integrate with Google services like Calendar and Tasks, allowing features such as adding dates from flyers or creating shopping lists from images. While these updates are set to arrive soon, Google has not provided a specific timeline for their release.

Cohere Boosts AI Fine-Tuning Efficiency

Cohere has updated its AI fine-tuning service, making it easier for businesses to create custom language models. The enhancements include real-time training monitoring, increased token capacity, and better customization for specific tasks using the Command R 08-2024 model. This model delivers faster performance with lower costs, ideal for industries with specialized needs like healthcare or finance. 

Integrating with Weights & Biases allows enterprises to track model fine-tuning and optimize performance. As competition in the AI industry intensifies, Cohere’s focus on customization and efficiency aims to attract businesses with unique language processing demands.

🧠RESEARCH

RATIONALYST: Pre-training Process-Supervision for Improving Reasoning

RATIONALYST is a model designed to improve reasoning by pre-training on rationale annotations. It extracts 79,000 rationales from large datasets and fine-tunes an existing model, LLaMa-3-8B. RATIONALYST improves reasoning accuracy by 3.9% across various tasks, outperforming larger models like GPT-4.

PHI-S: Distribution Balancing for Label-Free Multi-Teacher Distillation

PHI-S is a method for improving student models in multi-teacher distillation without labels. It standardizes teacher models’ activation statistics using Hadamard matrices, ensuring balanced distributions across dimensions. This technique, called PHI Standardization, enhances student model performance by better aligning teacher outputs, producing superior results compared to other methods studied.

TPI-LLM: Serving 70B-scale LLMs Efficiently on Low-resource Edge Devices

TPI-LLM is a system for efficiently running 70-billion-parameter models on low-resource edge devices. It optimizes memory and computation using tensor parallelism, keeping sensitive data local. TPI-LLM significantly reduces inference time and memory usage by managing model layers dynamically and improving communication via a star-based algorithm, outperforming existing solutions by over 80%.

Atlas-Chat: Adapting Large Language Models for Low-Resource Moroccan Arabic Dialect

Atlas-Chat is the first language models specifically designed for Moroccan Arabic (Darija). Using a combination of existing resources and new datasets, Atlas-Chat-9B and 2B models outperform other Arabic-focused models like LLaMa and Jais by 13% on Darija tasks. The study also explores optimal fine-tuning strategies for low-resource languages, making the models publicly accessible.

LEOPARD : A Vision Language Model For Text-Rich Multi-Image Tasks

LEOPARD is a vision-language model tailored for tasks involving multiple text-rich images, such as presentation slides or scanned documents. It addresses challenges like the lack of instruction-tuning datasets and the difficulty in managing image resolution and sequence length. LEOPARD uses high-quality multimodal datasets and adaptive encoding to optimize image understanding, outperforming current models in text-rich multi-image tasks.

🛠️TOP TOOLS

Pika 1.5 - Stunning footage. Longer clips. Jaw-dropping moves.

Vox - An AI-voice agent indistinguishable from a human

CoFrame - Generate visually-aligned website sections with a single prompt.

Sembly AI - Outsource your time‑consuming tasks to AI

Hedy - Provides real-time, customized insights and recommendations to help you excel in your business meetings and classes.

📲SOCIAL MEDIA

🗞️MORE NEWS

• Apple has launched Depth Pro, an AI model that generates precise 3D depth maps from a single image without needing traditional camera data. This breakthrough enhances industries like augmented reality and autonomous vehicles by providing real-time, detailed spatial awareness. Depth Pro is open-source, promising broad application potential.

• Google has introduced new AI security features for Android, such as Theft Detection Lock, Offline Device Lock, and Remote Lock, aimed at preventing phone thefts. These features automatically lock the phone if it detects suspicious activity, ensuring better protection of user data, even if disconnected from the internet or stolen.

• Plumerai, backed by Tony Fadell, brings advanced on-device AI to home security cameras, enhancing privacy by processing data locally without relying on remote servers. The startup’s efficient AI enables features like people detection and facial recognition, offering cost-effective, accurate performance. Chamberlain Group will integrate Plumerai’s AI into its smart cameras.

• Researchers at Cleveland Clinic and IBM are using AI to find non-addictive, non-opioid pain relief options. Their deep-learning framework, LISA-CPI, predicts how gut metabolites and FDA-approved drugs interact with pain receptors, identifying potential alternatives for chronic pain treatment.

• Captions, an AI-powered video editing app, has launched a social media manager tool that automates content creation and scheduling for websites. It scans the site’s content to generate relevant videos, focusing on platforms like Instagram Reels and TikTok.

• Elon Musk's AI startup, xAI, has moved into OpenAI's former headquarters in San Francisco's Mission district. Despite Musk's plans to relocate his companies to Texas, xAI continues to operate from both its new office and its Palo Alto location. The move occurred shortly after OpenAI shifted to a nearby office.