🤝 AidfulAI Newsletter #6: Express Yourself with AI
Dear curious minds,
Before diving into this issue, allow me to share a quick tip on how to enjoy the newsletter. Scan the titles and images and see what catches your eye. You can read it right away or save it for later in your favorite app, such as Instapaper or Readwise Reader.
Major AI News
The first two news in this section were already released shortly before the last issue of the newsletter. However, as from my current perspective, not much interesting from an end user perspective happened this week, I will cover them now.
🔮📸 Future of Image Editing: Adobe's Generative AI Revolution in Photoshop
Adobe has released a major update to Photoshop that integrates Adobe Firefly into the popular image editing software. Firefly is a family of generative AI models that create professional quality content from simple text prompts. Photoshop users can now access them via a feature named Generative Fill, to add, remove, or replace content in their images non-destructively. Generative Fill is available in every selection tool in Photoshop and produces a new generative layer type that can be edited with Photoshop's comprehensive range of tools.
If you are not a subscriber of Photoshop or the complete Creative Cloud, you can test it free for seven days (tip: cancel it right after you sign up). To use and explore the new generative fill feature, you have as of today to install the beta version of Photoshop.
🖥️🎯 Aiming Higher: Microsoft's AI Assistants Family Welcomes Windows Copilot
Microsoft has announced the next version of its AI assistants family named Copilot. After the overwhelmingly positive user reception of the AI pair programmer GitHub Copilot, Microsoft 365 Copilot was announced in March, intending to bring Large Language Models (LLMs) to their office package including calendar, emails, chats, meetings, documents, and more.
Now, the next step has been made public with the announcement of Windows Copilot, a new feature coming to Windows 11 that aims to boost user productivity. As a centralized AI assistant, it's easily accessible from the taskbar and operates in a global sidebar across all your apps and programs. It can handle a variety of tasks, from changing Windows settings, opening the relevant programs for your questions with specific settings to leverage AI to rewrite, summarize, or explain your content. To get a first glimpse of it, you can watch the video linked in the blog article announcing the new feature. Windows Copilot is scheduled for a preview release in June.
📝💼 Making Meetings Smarter: Microsoft Launches Intelligent Recap in Teams Premium
Microsoft has unveiled the general availability of Intelligent Recap, a new feature in Teams Premium. This tool utilizes GPT, developed by OpenAI, to provide users with personalized meeting summaries, tasks, chapters, and highlights. With this feature, users can easily access meeting transcripts, chat logs, shared content, and Loop notes created during the meeting.
Intelligent Recap in Microsoft Teams Premium enables users to review important details or catch up on missed meetings effortlessly. As the name suggests, this feature is not included in the standard Microsoft 365 subscription. Microsoft Teams Premium is priced at $10 per person per month, but a 30% discount is currently being offered.
This issue covers open-source LLMs in this section because they let users keep their data on their servers and apply techniques to safeguard sensitive information. Furthermore, users can see exactly how the model works and what data it has been trained on. This makes it easier to identify any potential privacy risks.
🤗🏆 Open LLM Leaderboard: A platform to rank and evaluate open-source LLMs and chatbots
The Open Large Language Model (LLM) Leaderboard is a new platform from Hugging Face which is designed to track, rank, and evaluate open-source LLMs and chatbots as they are released. This leaderboard allows anyone from the community to submit a model for automated evaluation, provided it is a Transformers model with weights on the Hub. The platform also supports evaluation of models with delta-weights for non-commercial licensed models, such as LLaMA from Meta.
The leaderboard evaluates models on four key benchmarks from the Eleuther AI Language Model Evaluation Harness. These benchmarks include the AI2 Reasoning Challenge, HellaSwag, MMLU, and TruthfulQA, all designed to measure a model's reasoning and general knowledge across a wide range of fields. The goal of the Open LLM Leaderboard is to filter genuine progress made by the open-source community and identify the current state-of-the-art models in the field.
🦅🌎 Falcon 40B: Abu Dhabi's Breakthrough in AI Language Models
A new artificial intelligence model developed by the Technology Innovation Institute (TII) in Abu Dhabi has taken the top spot on the Hugging Face leaderboard (described above).
Falcon 40B is a foundational large language model (LLM) with 40 billion parameters trained on one trillion tokens. It can generate natural language for various applications such as chatbots, engineering, healthcare, sustainability, and coding. Unlike most LLMs, Falcon 40B is fully open-sourced for research and commercial use, making it accessible to anyone who wants to explore its capabilities. Falcon 40B is a remarkable achievement that showcases Abu Dhabi's rapid development in the technology industry and its contribution to the global AI community.
PKM and AI
This section continues the exploration of how AI can be used in applying the CODE framework from Tiago Forte.
📚💡 AI and the Art of Book Summarization by Tiago Forte
Tiago Forte tested the newest model, GPT-4, in ChatGPT for creating book summaries. Since the AI in ChatGPT was only trained with content up to September 2021, and he intended to test the AI's ability to create summaries without any input from his notes, he chose a book published before this cut-off. The selected book was “Slack: Getting Past Burnout, Busywork, and the Myth of Total Efficiency” by Tom DeMarco.
In his blog article, Tiago shares what he learned from this experiment with ChatGPT. He found that if you provide the AI with carefully selected parts from a book, organized in a structured manner, it can produce a better summary than without any additional input. He points out that the main challenge is ensuring these parts are concise enough to fit within the AI's context window, which he estimates at 1,000 to 1,200 words. Small correction from me: The size limitation is currently around 2,400 words for GPT-4 and 4,800 words for GPT-3.5 in ChatGPT. If that's not sufficient, you might consider using the competing model, Claude from Anthropic, which currently supports up to 75,000 words.
Tiago states that it's challenging to prompt the AI to explain things in a new or unexpected way. The AI tends to stick to common sayings and typical ways of expressing ideas, so you need to creatively phrase your questions to prompt it to respond differently. Despite the advancements in AI, Tiago emphasizes the importance of reading books personally. When you read a book, you gain special insights and details the AI can't provide. By creating your own highlights and notes while reading, you capture what is important to you. Your notes represent your unique understanding of a book, which no one else can replicate. In a world where AI can rapidly summarize any content, having unique notes to create a personalized book summary gives you a competitive edge.
⚖️📝 Express Yourself with AI: Balancing AI Assistance and Human Touch in Content Creation
The revolutionary CODE framework by Tiago Fortes has found a powerful ally in AI. The true potential for getting AI assistance by applying CODE is the last stage “Express” where the new capabilities of creating and optimizing text can be used in countless ways.
Imagine having a series of notes, perhaps in the form of bullet points, that you want to transform into a well-articulated article. AI can help with that. Do you have a passage that needs truncation or expansion, or perhaps the inclusion of another relevant aspect? AI is at your service. From correcting spelling and grammatical errors to simplifying complex text, generating engaging titles, and even creating my emoji-led headlines – the possibilities with AI are truly boundless.
However, these advantages must not cloud the potential misuse of AI in creating massive amounts of low-quality content. It is crucial to remember that AI should serve as an assistant, aiding us in expressing our thoughts accurately, rather than being a factory for low-quality content.
As a reader of mine reminded my in a reply to the latest issue of this newsletter: To maintain transparency in this digital age, it's a good practice to include a disclaimer whenever AI has assisted in the content creation process. This simple gesture, which I implemented below, promotes accountability and generates trust among readers.
💼🎨 How AI is Changing Work and Creativity with Kevin Scott, CTO of Microsoft
They explore how Microsoft is pushing the boundaries of AI with GitHub Co-Pilot and ChatGPT. They also cover scaling up AI, the future of work and human creativity. This is a fascinating conversation with an influential leader in AI.
Thanks for reading! If you are not already subscribed, enter your mail address below to receive new issues and support my work.
Disclaimer: This newsletter is written with the help of AI. I use AI as an assistant to generate and optimize the text. However, the amount of AI used varies depending on the topic and the content. I always curate and edit the text myself to ensure quality and accuracy. The opinions and views expressed in this newsletter are my own and do not necessarily reflect those of the sources or the AI models.