GPT-4 Turbo Now Free in Microsoft Copilot

Sponsored by

Good morning. It’s Wednesday, January 17th.

Did you know: 40 years ago today, the U.S. Supreme Court ruled in a 5-4 decision that the personal taping of television programs using a home VCR to watch later is considered fair use and is not copyright infringement. It was known as the "Betamax case."

In today’s email:

  • Advancements and Innovations in AI

  • AI in Business and Partnerships

  • AI and Global Impact

  • 6 New AI Tools

  • Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

In partnership with MIT

Artificial Intelligence online short course from MIT

Study artificial intelligence and gain the knowledge to support its integration into your organization. If you're looking to gain a competitive edge in today's business world, then this artificial intelligence online course may be the perfect option for you.

  • Key AI management and leadership insights to support informed, strategic decision making.

  • A practical grounding in AI and its business applications, helping you to transform your organization into a future-forward business.

  • A road map for the strategic implementation of AI technologies in a business context.

Today’s trending AI news stories

Advancements and Innovations in AI

> Microsoft has integrated GPT-4 Turbo into its Copilot app, a move that cleverly sidesteps the $20 monthly fee that OpenAI’s ChatGPT typically commands. The updated Copilot, accessible on both iOS and Android, now boasts enhanced functionalities such as improved accuracy in responses and advanced image analysis, powered by OpenAI’s DALL-E 3 technology.

> Tencent ARC Lab’s Photomaker, an AI-driven photo generation tool, marks a significant advancement in personalized image creation. Lauded by AI experts like Yann LeCun, its ‘Stacked ID Embedding’ allows flexible integration of features from multiple IDs, enabling the creation of unique photos.

> ByteDance introduces MagicVideo-V2, an advanced text-to-video AI model surpassing others like Runways Gen-2. Combining text-to-image, image-to-video, video-to-video, and video frame interpolation modules, it produces high-quality, detailed videos, preferred by users in blind tests.

> Adobe is innovating in Premiere Pro by integrating AI-driven audio editing capabilities, designed to simplify and expedite tasks such as dialogue enhancement and tool navigation. The latest beta release introduces user-friendly features like interactive fade handles for rapid audio transitions, alongside an easily accessible Enhanced Speech function.

AI in Business and Partnerships

> Salesforce has introduced a new range of generative AI tools, aimed at enhancing productivity and conversion rates for retailers. Announced at the National Retail Federation’s Big Show, these tools are part of Salesforce’s Einstein 1 platform, integrating large language models with retail data for creating generative apps and automation. The lineup includes the Page Designer for building web pages using text prompts, Return Insights Tool for analyzing returns data to reduce future returns, and Inventory Insights for real-time inventory management. These tools, set to release in summer 2024, are designed to provide retailers with actionable insights and efficient ways to utilize emerging AI technology.

> A PwC survey indicates that 25% of CEOs globally expect to reduce their workforce by at least 5% this year due to generative AI. The survey, encompassing over 4,700 CEOs from 105 countries, reveals that a third have already integrated generative AI into their operations. While some job cuts are anticipated, particularly in media, entertainment, banking, and insurance sectors, there is also a potential for new hiring in other areas. Around 70% of CEOs foresee AI changing their business models and necessitating new skills within three years.

AI and Global Impact

> Jeffrey Katzenberg, co-founder of DreamWorks SKG, predicts that AI will replace 90% of artist jobs in animated film production. In his remarks on a Bloomberg TV panel, Katzenberg highlighted AI’s profound impact on media and entertainment, comparing it to revolutionary tools like the pen and camera. He reminisced about the days when 500 artists over five years were needed for top animated movies, a number he believes will drop to just 10% in three years. Katzenberg emphasizes that while AI will dramatically change production, individual creativity will still pay a crucial role in guiding AI, turning “prompting” into a valuable skill in the industry.

> Researchers in Abu Dhabi have developed an AI capable of mimicking handwriting, potentially assisting individuals unable to write but also raising concerns about mass forgeries. The AI, learning from a few paragraphs of handwritten text via a transformer model, has earned a U.S. patent. This innovation from Abu Dhabi offers benefits for the injured but also poses significant risks related to forgery and authenticity.

> Stability AI introduces 'stable-code-3b', a powerful 2.7B parameter language model pre-trained on 1.3 trillion tokens, including code datasets. Designed to excel across 18 programming languages, the model showcases superior performance on the MultiPL-E metrics. This model, equipped with innovative features like 'Fill in Middle Capability', is poised to revolutionize code generation, offering extensive language support and state-of-the-art architectural modifications for enhanced performance and efficiency.

6 new AI-powered tools from around the web

AI Picasso is a video creation app that transforms photos into full-body dance videos. This app offers long duration and realistic animations for Tik-Tok, YouTube Shorts, and more. offers AI voice-over studio with 400 voices in 50 languages and 8,000 soundtracks. It features real voice cloning, OpenAI integrations for tailored recordings, serving both businesses and individuals in diverse applications.

Friz is an AI-driven social media manager designed for D2C brands, streamlining content creation across platforms. It assists in brainstorming, editing, and scheduling posts, with a focus on contextual visuals and captions, all from a single tool.

Pinokio is an AI browser that allows users to install, run, and control any application programmatically and automatically, catering to both technical and non-technical users, with its open-source, user-friendly interface.

MakeShorts AI streamlines short-form video creation by transforming YouTube content into captivating shorts with AI Clipping, Autofocus Cropping, and AI Subtitling. It features a combat writer’s block and offers efficient, pay-per-minute editing.

AdIntelli simplifies monetization for GPT creators by integrating in-chat ads, offering a no-code solution to turn your GPTs into revenue streams with minimal effort and enhanced user experiences.

arXiv is a free online library where researchers share pre-publication papers.

The paper introduces Activation Beacon, a novel method enhancing LLMs like Llama-2-7B to process up to 400K context lengths. This plug-and-play module condenses raw activations into compact forms, enabling efficient processing of extended contexts while preserving short context capabilities. It employs auto-regression, trained efficiently with short sequences, outperforming existing methods in long-context tasks. Its innovative approach promises a low-cost, effective, and compatible solution for extending LLM context lengths.

AST-T5 is an innovative approach in code generation and understanding, developed by researchers from UC Berkeley and Meta AI. It stands out by utilizing Abstract Syntax Trees (ASTs) in a new pretraining paradigm, significantly enhancing the model's ability to process code. This technique ensures the preservation of code structure through dynamic programming and employs an AST-Aware Span Corruption objective, enabling effective reconstruction of diverse code structures. Remarkably, AST-T5 integrates easily with any encoder-decoder Transformer model, avoiding the need for complex program analyses or architectural changes. Its performance surpasses similar-sized language models in various code-related tasks, especially in code-to-code tasks, demonstrating a clear advantage over models like CodeT5 in specific benchmarks like Bugs2Fix and Java-C# Transpilation.

The paper "Infinite-LLM: Efficient LLM Service for Long Context with DistAttention and Distributed KVCache" by Alibaba Group and Shanghai Jiao Tong University introduces DistAttention, an algorithm enabling distributed processing and storage of attention modules for Large Language Models (LLMs) on the cloud. It segments Key-Value (KV) Cache into smaller units (rBlocks) for efficient handling of long contexts. DistKV-LLM, a distributed LLM serving system, is proposed to manage KV Cache dynamically across GPU and CPU memories in a data center. This approach allows handling context lengths up to 1,900K, showcasing 1.03-2.4× throughput improvements over existing systems in a 32 NVIDIA A100 GPU environment.

Gather-Attend-Scatter (GATS) by Google DeepMind introduces a versatile module for AI systems, enabling seamless integration of various pre-trained foundation models into a larger multimodal network. GATS facilitates the combination of trainable and frozen models, supporting diverse data rates and modalities, and preserving original model knowledge by avoiding traditional fine-tuning. Demonstrated across various domains like gaming and robotics, GATS exhibits significant potential in advancing AI by efficiently amalgamating multiple modalities without necessitating component model alterations.

The research project, conducted at Cornell University, focuses on employing machine learning to differentiate AI-generated synthetic images from real ones, addressing the growing challenge of distinguishing authentic from fabricated visual content in digital media. The study, leveraging the CIFAKE dataset, involves refining and adapting advanced deep learning architectures like ResNet, VGGNet, and DenseNet, and comparing them to baseline models. The project, notable for its high accuracy rates, particularly with DenseNet, contributes significantly to the field of digital media integrity and paves the way for future research into the ethical and technical aspects of AI-generated content.

ChatGPT Creates Comics

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, apply here.