Replit Agent, world’s top open-source model, new real-time audio conversational model, AlphaProteo, style vs substance, fully open-source mixture-of-expert (MoE) language model and more
Replit Agent, world’s top open-source model, new real-time audio conversational model, AlphaProteo, style vs substance, fully open-source mixture-of-expert (MoE) language model and more
Hi. Welcome to this week's AI Brews for a concise roundup of the week's major developments in AI.
In today’s issue (Issue #75 ):
AI Pulse: Weekly News & Insights at a Glance
AI Toolbox: Product Picks of the Week
From our partners:
Screen Studio: Beautiful Screen Recordings in Minutes
Screen Studio is an simple yet powerful screen recorder that makes your videos look beautiful. It automatically zooms in on your cursor, increases cursor size and smooths its movements. Use it to record and create engaging product demos, courses, tutorials and social media posts
🗞️🗞️ AI Pulse: Weekly News & Insights at a Glance
🔥 News
Replit launched Replit Agent, an AI agent for building fully-functional apps using natural language. It automates the entire app development process, including setting up environments, managing dependencies, configuring databases, and deploying apps to the cloud. Replit Agent is available today in early access to all core subscribers [Details]
Ai2 and Contextual AI released OLMoE, a first-of-its-kind fully open-source mixture-of-expert (MoE) language model with 1 billion active and 7 billion total parameters that that beats comparable LLMs and can be run easily on common edge devices. OLMoE is pre-trained from scratch and released with open data, code, logs, and intermediate training checkpoints [Details].
Matt Shumer, co-founder and CEO of AI HyperWrite, released Reflection 70B, a new model trained from Llama 3.1 70B Instruct, claiming it to be the world’s top open-source model. It beats GPT-4o on every benchmark tested. It is trained using Reflection-Tuning, a technique developed to enable LLMs to fix their own mistakes. All benchmarks tested have been checked for contamination by running LMSys's LLM Decontaminator [Details | HuggingFace].
Researchers released Mini-Omni, an open-source multimodel large language model that can hear, talk while thinking (the ability to generate text and audio at the same time). It features real-time speech-to-speech conversational and streaming audio output conversational capabilities [Details].
Luma AI released Dream Machine 1.6 with Camera Motion that lets you direct text-to-video and image-to-video scenes with simple commands [Video]
Function Calling is now available in Google AI Studio making it easy to test the models capability quickly without leaving the UI [Link].
LMSYS developed a method to understand the effect of style vs substance when evaluating chatbot models in the Chatbot Arena and found noticeable shifts in the ranking. GPT-4o-mini and Grok-2-mini drop below most frontier models, and Claude 3.5 Sonnet, Opus, and Llama-3.1-405B rise substantially [Details]
Chinese start-up MiniMax funded by Alibaba launched video-01, its new text-to-video-generating model. It’s available on its consumer-facing Hailuo AI platform [Details].
Cohere released improved versions of Command R and Command R+, the enterprise-grade AI models optimized for business use-cases [Details].
01.AI released Yi-Coder 1.5B and 9B open-source code LLMs under Apache 2.0. Yi-Coder-9B outperforms other models with under 10 billion parameters, such as CodeQwen1.5 7B and CodeGeex4 9B, and even achieves performance on par with DeepSeek-Coder 33B [Details].
Google DeepMind introduced AlphaProteo: an AI system for designing novel proteins that bind more successfully to target molecules. It could help scientists better understand how biological systems function, save time in research and advance drug design [Details].
Hacker Cup, Meta's annual open programming competition, adds AI division for 2024 season in which all of your solutions must be written by a computer. First round will start on September 20th [Details].
Microsoft gives deepfake porn victims a tool to scrub images from Bing search [Details].
Hiring platform ZipRecruiter is launching a new AI-powered tool, called ZipIntro, to let employers match and schedule introductory calls with potential candidates [Details].
Anthropic is launching a new subscription plan for its AI chatbot, Claude, catered toward enterprise customers with larger context window and GitHub integration [Details].
Google AI Overviews rollout hits news publisher search visibility. AI Overviews now being offered for 17% of queries in UK and US [Details].
LLaVA v1.5 7B , a powerful visual model, is now available on GroqCloud Developer Console [Details].
🔦 Weekly Spotlight
StreamingT2V: generate high-quality, long videos with rich motion dynamics [Link].
Anthropic Quickstarts: a collection of projects designed to help developers quickly get started with building deployable applications using the Anthropic API [Link].
Video-to-Video for CogVideo: take any video and turn it into another video [Link].
Speech To Speech: an effort for an open-sourced and modular GPT4-o [Link].
HivisionIDPhoto: ID photo production framework. Uses a set of models and workflows for portrait recognition, image cutout & ID photo generation for a variety of photography situations [Link].
🔍 🛠️ AI Toolbox: Product Picks of the Week
QuickMagic: Turn your videos into animations with ease and create high-precision motion graphics.
UseFlux: Train high-quality AI image models for various purposes, including headshots, portraits, products, and brand assets using Flux AI image model.
Melodio: Melodio can stream personalized music for over 10 hours. Melodio AI uses its own proprietary music model.
AI Ads Analyzer by GoMarble: AI-powered tool for creative analysis of any video or static ad.
Last week’s issue
You can support my work via BuyMeaCoffee.
Thanks for reading and have a nice weekend! 🎉 Mariam.