Truly Open Models, Code Llama 70B, Amazon AI Hackathon , AI Grant, world’s greenest 7B model and more
Greetings and welcome to this week's AI Brews for a concise roundup of the week's major developments in AI.
In today’s issue (Issue #50 ):
AI Pulse: Weekly News & Insights at a Glance
AI Toolbox: Product Picks of the Week
AI Skillset: Learn & Build
🗞️🗞️ AI Pulse: Weekly News & Insights at a Glance
🔥 News
Allen Institute for AI (AI2) releases Open Language Model (OLMo), a series of truly open language models and framework, with the goal to collaboratively build the best open language model. OLMo releases the whole framework from data to training to evaluation tools: multiple training checkpoints across multiple hardware types, training logs, and exact datasets used, with a permissive license. The OLMo models (7B/1B) are trained on the Dolma dataset [Details | Hugging Face].
Meta AI introduces and open-sources AudioSeal, the first audio watermarking technique designed specifically for localized detection of AI-generated speech [Details].
Nomic AI released Nomic Embed, the first fully-open text embedding model with a 8192 context-length that outperforms OpenAI Ada-002 and text-embedding-3-small on both short and long context tasks [Details].
Nvidia and Suno present Parakeet-TDT, a state of the art English Automatic Speech Recognition (ASR) model with the latest Token and Duration Transducer. Parakeet TDT achieves unrivaled accuracy while running 64% faster over their previous best model [Details | Hugging Face] .
Meta AI released Code Llama 70B, an open-source model based on Llama 2 for code generation. CodeLlama-70B-Instruct achieved 67.8 on HumanEval, reaching the initial GPT-4 performance. CodeLlama-70B-Instruct is now on Perplexity Labs and PPLX-API. [Details | Hugging Face].
Alibaba Cloud presents Qwen-VL-Max, an upgraded version of their open sourced multimodal model Qwen-VL. It offers a higher level of visual perception and cognitive understanding and performs on par with Gemini Ultra and GPT-4V in multiple text-image multimodal tasks, significantly surpassing the previous best results from open-source models [Details | Hugging Face].
RWKV released Eagle 7B, a 7.52B parameter multi-lingual model built on the RWKV-v5 architecture, allowing 10-100x+ lower inference cost. Trained on 1.1 Trillion Tokens across 100+ languages, it outperforms all 7B class models in multi-lingual benchmarks and ranks as the world’s greenest 7B model (per token). RWKV-v5 Eagle 7B can be used personally or commercially without restrictions [Details]
Researchers released LLaVA-1.6 large multimodal model that has improved reasoning, OCR, and world knowledge. LLaVA-1.6-34B outperforms Gemini Pro on benchmarks like MMMU and MathVista [Details].
Google’s Bard, powered by the Gemini Pro, surpasses GPT-4 to the #2 position on ‘LMSYS Chatbot Arena Leaderboard’, the crowdsourced open platform for LLM evals [Link].
Mistral CEO confirms ‘leak’ of new open source AI model (miqu-1-70b) nearing GPT-4 performance, on Hugging Face and on 4chan [Details].
The dataset used to create Open Hermes 2.5 and Nous-Hermes 2 is now public [Details].
Google Research introduced MobileDiffusion, an efficient latent diffusion model specifically designed for text-to-image generation on mobile devices. It can run in half a second to generate a 512x512 high-quality image on iOS and Android premium devices with small model size being just 520M parameters [Details].
Paid users of ChatGPT, OpenAI’s AI chatbot front end, can bring GPTs into a conversation by typing “@” and selecting a GPT from the list. The chosen GPT will have an understanding of the full conversation, and different GPTs can be “tagged in” for different use cases and needs [Details].
Amazon Web Services announced PartyRock Generative AI Hackathon with $120,000 in prizes (AWS credit + cash). No coding experience is required. Builders will need to build a generative AI app using PartyRock, a shareable generative AI app building playground with a web-based UI [Details].
AIWaves introduced Weaver, a family of large language models (LLMs) dedicated to content creation. It includes Weaver Mini (1.8B), Weaver Base (6B), Weaver Pro (14B), and Weaver Ultra (34B) sizes, suitable for different applications and can be dynamically dispatched by a routing agent according to query complexity to balance response quality and computation cost. Weaver Ultra model surpasses GPT-4, on various writing scenarios [Paper].
Apple Podcasts is getting auto-generated transcripts with iOS 17.4 [Link].
Elon Musk’s neurotech startup Neuralink implanted its device in a human for the first time, and the patient is “recovering well. The brain implant aims to help patients with severe paralysis control external technologies using only neural signals [Details].
Batch 3 application for AI Grant are now open and will close on February 16 [Details].
Abacus AI introduced SMAUG, a 30B open-source model achieving an MMLU of 76.66 , outperforming other open-source models in the 30B class [Details]
Shopify is rolling out an AI-powered image editor for products [Details].
The New York Times is building a team to explore AI in the newsroom [Details].
🔦 Weekly Spotlight
Enchanted, an open source, Ollama compatible, iOS app for chatting with privately hosted models such as Llama 2, Mistral, Vicuna, Starling and more [Link].
The promise and challenges of crypto + AI applications by Vitalik Buterin [Link].
‘AI Opportunity Agenda for ASEAN', a whitepaper by Google to help ASEAN governments tap into AI’s vast potential [Link].
How enterprises are using open source LLMs: 16 examples [Link].
WhisperKit, a Swift package that integrates OpenAI's Whisper speech recognition model with Apple's CoreML framework for efficient, local inference on Apple devices [Link]
🔍 🛠️ AI Toolbox: Product Picks of the Week
Arc Search: An iOS app from the company behind the Arc browser for AI-enabled mobile browsing. Browse for Me feature reads multiple web pages, and creates a custom web page for you.
Promptly: a no-code Generative AI platform for advanced app and workflow creation. Choose from popular LLMs, add data in any format, and use AI Agents for tasks from data retrieval to online form completion.
📕 📚 AI Skillset: Learn & Build
Prompting Guide for Code Llama [Link].
How AI Works - An entirely non-technical explanation of LLMs [Link].
Thanks for reading and have a nice weekend! 🎉 Mariam.