🤝 AidfulAI Newsletter #14: Unveiling Stable Diffusion XL 1.0
Dear curious minds,
Today’s issue brings you exciting updates on ChatGPT's new Beta feature, the mobile-friendly BTLM-3B-8K, and the newly launched Stable Diffusion XL 1.0. Plus, note-taking prompts and fascinating podcast discussions. Let's dive in!
Major AI News
📝🔧 ChatGPT: New Beta feature "Custom instructions"
It can be activated for Plus subscribers in the settings.
Allows defining your own system prompt.
Split into “about you” and “how to respond” sections with up to 1500 characters each.
More in the official announcement article.
Pro tip: Give ChatGPT the instruction to never answer with "As a large language model ...". For me, this only worked in GPT-4. Based on all the attention this tip got in the tweet from Jeremy Nguyen, OpenAI already announced that this will not be needed anymore with the next ChatGPT release.
My take: The same has been possible via API for a long time, but we learned with ChatGPT how much difference a UI can make.
Privacy-Friendly AI
📱💬 AI on your phone: BTLM-3B-8K, a state-of-the-art language model that fits on your mobile device
BTLM-3B-8K: an open, state-of-the art 3B parameter model with 7B level performance.
When quantized, it fits in as little as 3 GB of memory.
It runs on iPhone, Google Pixel, even Raspberry Pi.
BTLM goes is announced to go live later this week on Bittensor, a blockchain-based network that lets anyone contribute AI models for inference and provides a decentralized alternative to centralized model providers like OpenAI.
More info in the announcement blog article.
My take: This is not just an advancement, it's the future unfolding before our eyes.
🔥🚀 Stable Diffusion XL 1.0 released: A Midjourney competitor
The previous version, Stable Diffusion XL 0.9 was released a little more than a month ago on the 22nd of June.
In contrast to 0.9, the license of Stable Diffusion XL 1.0, does limit usage to research purposes.
Compared to the previous version 0.9, the upgrade includes:
more vibrant colors,
better contrast,
enhanced lighting and shadows,
faster results,
multiple aspect ratios.
Accessible via:
GitHub (weights and associated source code).
The official announcement also states that fine-tuning the model with your own data and guiding the generation process with image inputs will become easier than ever before and is currently in beta preview
My take: After quite some time, this looks like a respectable competitor to Midjourney.
PKM and AI
📝💡 Reflect App's Top AI Prompts for Note-Making
The note-taking app Reflect integrates GPT-4 from OpenAI for all users.
In a recent blog article, they shared their insights about the most useful prompts for note-taking. These are inbuilt in Reflect, but can be used manually in other AI tools, like ChatGPT or Claude. The following is only an overview, to get the actual prompts, go to the blog article:
List Key Takeaways: This prompt will take any notes you have and list the key takeaways.
List action items: This prompt will find the actionable steps from any selected text and list them out for you.
Act as your copy editor: This prompt will ensure your text is void of any spelling, grammar or punctuation mistakes. It will also improve general readability and flow.
Format voice notes into structured paragraphs: This prompt will format the text into nice, clean paragraphs.
Simplify and condense your writing: This prompt will turn your unorganized rambles into writing that anyone can easily understand.
Act as your Executive Assistant for meetings: This prompt will take your meeting notes and extract the key takeaways and action items from them.
Fill in missing information: This AI prompt will both find and fill in these gaps in your notes.
Automatically add backlinks: This prompt adds bi-directional backlinks in your note.
Top Priorities from a voice note: This custom prompt extracts tasks from a voice note.
Translate text into any language: Traditional language translators have been moderate at best in their outputs. AI is excellent at translating! Use this prompt and replace the language with whatever you need to translate your writing into.
My take: Beside the prompts themselves, these might add value by giving you some fresh ideas of how to take notes. Thanks to the transcription and summarization capabilities of AI, voice notes might become way more valuable.
Podcasts
🎙️🎬 How AI is reshaping journalism and Hollywood: A conversation with Emad Mostaque
Emad Mostaque, CEO and Co-founder of Stability AI, was interviewed by Peter Diamandis in the Moonshots and Mindsets podcast episode #55
They discuss the potential of AI to transform journalism and Hollywood, while also highlighting the challenges and risks associated with this technology.
Coders of the future will heavily rely on AI tools, resulting in a decrease in time spent on debugging and unit testing.
Providing better data to AI and developing a counter AI to prevent harmful AGI are possible solutions to reduce AI risk.
My take: I like how Emad envisions the future of generative AI, with each company and nation having their own LLMs.
🎙️🦙 Unveiling Llama 2: Insights from the Latent Space Podcast with HuggingFace and a16z Guests
The Latent Space Llama2 podcast was realized as a Twitter space four days after the Llama 2 release.
Guests in this podcast were Sebastian Borgeaud, a machine learning researcher at Huggingface, and Matt Bornstein, an organizer of the a16z infrastructure team that launched Llama2.ai where you can test and use the different model versions.
The community will get their hands on the model and understand its capabilities. There will be a period of fine-tuning and application to explore the model's potential.
My take: Releasing Llama 2 as an open-source model for commercial use is likely the best way for Meta to catch-up in the LLM race.
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.