🤝 OpenAI's Turbulent Week Underscores the Need for Alternatives
Dear curious minds,
Welcome to the ultimate newsletter for those interested in Artificial Intelligence (AI) and Personal Knowledge Management (PKM).
You might have observed small changes in this issue. I dropped the prefix “AidfulAI Newsletter #x:” in the headline and rebranded the newsletter to “Aidful News”. Furthermore, I updated the logo, which, like the one before, represents the collaboration with AI by a handshake between a human and a robot (created by DALL·E 3 and tweaked in GIMP).
In this week's issue, I bring to you the following topics:
The Turbulent Week at OpenAI: CEO Shuffles and Groundbreaking Discoveries
Unveiling Claude 2.1: Anthropic's Major Upgrades and New Features
Introducing Copilot in OneNote: Revolutionizing Note-Taking and Planning
Run Local LLMs: Experience the Lightning-Fast Power of GPT4all
Exploring Creativity and Productivity with Ali Abdaal and Nick Milo
If nothing sparks your interest, feel free to move on, otherwise, let us dive in!
🤖🌪️ The Turbulent Week at OpenAI: CEO Shuffles and Groundbreaking Discoveries
What happened last week at OpenAI will make it in the history books, it is just unbelievable how fast everything happened. The following chain of events started on Friday 17th of November:
Friday: OpenAI announces leadership transition. CEO Sam Altman was fired by the board and the CTO, Mira Murati, was announced as interim CEO. President Greg Brockman did step down from his position.
Sunday: Microsoft CEO Satya Nadella initiates negotiations between Sam Altman and the board. Emmett Shear, former CEO of the video game streaming service Twitch, was named as the new interim CEO of OpenAI.
Monday: Satya Nadella shared on 𝕏 that Sam Altman and Greg Brockman will lead a new AI research team at Microsoft. Over 700 of OpenAI’s 770 employees signed a letter to the board that they need to step down.
Wednesday: OpenAI announced on 𝕏 that Sam Altman will return as CEO with a new initial board.
Today, Reuters reports that OpenAI researchers warned their board about a potentially dangerous AI discovery, which played a significant role in CEO Sam Altman's firing. The concern was about the risks of commercializing this breakthrough without fully grasping its consequences.
My take: The rapid sequence of these events and the high stakes involved make the entire situation seem almost surreal. No one knows what was the reason for the board decision, and there are many speculations. My recap of the events did not cover all the details, and you need to dig a bit more for yourself if you are interested in getting a better understanding. Nevertheless, I am sure that this will not be the final chapter of OpenAI, and we will likely get more insights in the next couple of days, weeks and months about this week.
🤖📈 Unveiling Claude 2.1: Anthropic's Major Upgrades and New Features
Anthropic announced an upgraded version of their Claude chatbot to be used via API and at Claude.ai.
Claude 2.1 boasts an impressive 200K token context window. That’s roughly 150K words, or over 500 pages! This allows for processing entire codebases, as well as summarizing and chatting with long form content like books or podcasts.
They've achieved a 50% reduction in false statements with Claude 2.1 compared to the previous version.
Hallucinations are reduced significantly in Claude 2.1 Interesting to see the error differences in document position and total length.
Anthropic announced a beta tool feature which allows calling for arbitrary functions, web source searching, and retrieval from private knowledge bases. If you are interested, put yourself on the waitlist.
Development with Claude has been made easier with the introduction of a new Workbench feature for prompt testing, a simplified prompt iteration process, and system prompts for tailored instructions and structured responses.
The API pricing of Claude 2.1 and their other models was reduced.
The new pricing is effective from November 1, 2023. My take: These upgrades position Claude as a strong alternative for users needing to work with longer documents and larger file sizes. From my perspective, Claude is currently the best performing competitor to the models we can use in ChatGPT.
🤖✍️ Introducing Copilot in OneNote: Revolutionizing Note-Taking and Planning
The last issue highlighted the rebranding from AI used by Microsoft to Copilot. The paid version is named Copilot for Microsoft 365.
One of the tools in the Microsoft 365 package is the Windows desktop application OneNote, which now gets Copilot functionality.
The following is a summary of the announced features of Copilot in OneNote, demonstrating its utility in enhancing productivity and creativity in note-taking and planning:
Side pane: The Copilot pane can be accessed from the OneNote ribbon or via the Alt+H+FX keyboard shortcut, with the ability to enter custom prompts for varied responses.
Summaries: Copilot can summarize text notes, create to-do lists, list pros and cons, and rewrite notes for clarity.
To-Do Lists: It can generate to-do lists from OneNote selections, pages, and sections, helping to manage follow-up tasks during planning sessions.
The Copilot lives in the OneNote side pane and can generate new pages and add content to existing ones. Listing Pros and Cons: Offers insights by listing pros and cons, providing clarity and guidance on various subjects within notes.
Content Rewriting: Assists in rewriting content for tone, style, formality, and length, with options for different rewritten responses.
Idea Generation: Helps in overcoming creative blocks by providing ideas and inspiration for new content, plans, and presentations.
Planning Assistance: Drafts plans for projects, campaigns, and personal goals, aiding in the initial steps of planning.
Chat Functionality: Enables free-form chatting with Copilot to find information, get tips, and learn new information to enhance notes.
Furthermore, Microsoft shares a prompting guide as PDF. It is rather basic but nicely done and worth a look.
My take: I used OneNote for many years at work and as much as I love the tool, I really miss the ability to seamlessly link notes like it is possible in Logseq or Obsidian. The new capabilities look very promising, but the use of Copilot, which leverages LLMs, inherently means that data is processed and potentially stored outside of one’s personal environment. This could be a significant issue for those handling sensitive or confidential information.
🌋🏃 Run Local LLMs: Experience the Lightning-Fast Power of GPT4all
GPT4all, which was already covered in issue #9, is a platform which allows running LLMs locally on everyday hardware.
Previously it was only fast on Macs with an M chip, as it uses on all other platforms only the CPU. Now, GPT4all can utilize any GPU (blog article).
Thanks to Vulkan GPU API, LLM inference is speeded-up on AMD, Intel, Samsung, Qualcomm, and NVIDIA GPUs.
The UI of GPT4all shows the generation speed in the lower right while it creates output. There are no downtimes or speed limitations like many have experienced in the last few weeks by using online systems like ChatGPT and its competitors.
My take: I am thrilled by the mission of bringing local large language models (LLMs) to the masses. The new GPU support is a fantastic feature, as it allows a wide range of systems to benefit from it. With my NVIDIA GeForce 3090 GPU, I'm now achieving responses faster than in ChatGPT, running models with 8 billion parameters at an impressive speed of 50 tokens per second.
🎙️🌟 Exploring Creativity and Productivity with Ali Abdaal and Nick Milo
Ali Abdaal, a renowned productivity expert, joins Nick Milo on the first episode of his new podcast “How I think” and shares his journey, including the creation of a successful YouTube channel.
Reflecting on journaling prompts like "What would you do if you couldn't fail?" and "What would you do even if you knew you would fail?" can help us assess our alignment with our desires and discover what we truly want in life.
They discuss impactful books and the evolving landscape of the creator economy, delving into the future of content creation and its possibilities.
Ali's new book 'Feel Good Productivity' is mentioned, which has a really nice landing page and offers a pre-order bonus in the form of an exclusive annual planning workshop in January.
My take: Listening to this insightful episode of "How I Think" with Ali Abdaal offers a unique opportunity to dive into the mind of two productivity gurus. From understanding the power of journaling to exploring the evolving creator economy, this podcast is a must-listen for anyone interested in harnessing their creative potential and discovering new paths to success.
Disclaimer: This newsletter is written with the aid 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.