Learn and Do with Peter, 2024-03-24

Recorded live AI Salon / Learn and Do with Peter session on AI topics.

YouTube: https://youtu.be/VwvIn-64yP4

Next: Learn and Do with Peter, 2024-03-29

Previous: Learn and Do with Peter, 2024-03-16

Notes: Learn and Do with Peter Notes, 2024-03-24

AI Summary (hopefully useful, may be inaccurate)

Quick Recap

Pete and Scott discussed their upcoming work tasks, including presentations, photo editing, and the use of Python and AI for various tasks. They explored the potential of using prompts in image generation and the Kaminski method, which combines two prompts to create an image. The team also discussed the implementation of an AI companion for note-taking during meetings, modifications to their AI image generation system, and ways to enhance their tool using randomization and documentation. Lastly, they focused on the technical aspects of their project, the challenges of controlling output randomness, and the importance of adding more pages to their website and documenting their work.


Planning Upcoming Tasks and Photo Editing Tools

Pete and Scott planned their upcoming work tasks, including a presentation on Obsidian for Ann Murphy's team and website editing to improve Scott's skills. They also discussed various photo editing tools, with Pete sharing his experiences and techniques in painting and digital art, including his experimentation with combining image prompts using chat GPT. The team's discussion highlighted the importance of effectively utilizing prompts in these processes.

Exploring Python Tools and Interleaver Application

Pete, Daniele, and R discussed the potential and implementation of using Python for various tasks, including working with files, text, and web applications. They explored a specific Python tool, Dragon, which they found on Github, with plans to experiment with it further. Additionally, Pete presented a new interleaver application developed in HTML, which can be easily distributed and run, with the aim to enable users to instruct Chachb to write a program in HTML rather than Python.

Image Generation Tool and Prompts

Pete shared his idea of creating interesting prompts for image generation using a new tool. He demonstrated how he uses the tool, including brainstorming sessions and generating various types of images. He also showed examples of his work, including a medieval mountain marketplace and a historical abstract landscape fantasy. Pete then guided Daniele on how to use the tool, with Daniele following his steps. The team planned to post their images for review later.

Kaminski Method and AI Integration

Pete introduced a new method called the Kaminski method, which combines two prompts to create an image. He shared his personal experiences with using technical tools as an artist and his experimentation with digital synthesizers for creating unique sounds. The team discussed the efficiency of letting machines generate a lot of content for review and the challenges of eliciting unique descriptions from language models. Pete also shared his plans to develop an app using prompts for creative writing and addressed a question from Daniele about his TikTok video. The team briefly discussed some user interface issues with the current system.

AI Image Generation and Prompt Writing

Pete proposed turning their discussed item into an open source project and addressed R's question about organizing information, particularly in relation to image generation instructions. They discussed their experiences with AI image generation, focusing on techniques such as describing scenes instead of individuals, and the challenges and results of sophisticated prompt writing. The team also explored the idea of adapting HTML features and incorporating photography terminology and artistic elements into prompts to guide the model. They agreed to continue the discussion on this topic in the future.

Exploring Photo Techniques and Javascript

Pete and R discussed the potential of using a library of photo techniques to create prompts for image creation. Pete suggested using a simple method of transforming text to create images, like interleaving or swapping adjacent characters, and generating an image by selecting random words from a large list. They also discussed the limitations and capabilities of Javascript within an HTML page, with Pete highlighting its simplicity and power, and planning to demonstrate a simple task using it.

AI Companion for Meeting Notes

Pete and R discussed the implementation of an AI companion to take notes during their meetings. Pete suggested several prompts to try, including interleave 3 prompts, reverse pairs of words, adding random comments, and adding random emojis. They decided to try the first pair of words. They also discussed creating an app called Reverse Mix, which would reverse the input text. They ended the session with an open question about what the AI should do with odd numbers of words.

Pete's Code Process and Debugging Demo

Pete demonstrated a process involving code and text files, including copying, pasting, and editing. He emphasized the importance of adding a copyright statement and a license, preferably a Mit license for software, and the value of having a checklist for release procedures. He also showed how to debug code using the built-in console in web browsers. Pete suggested that team members build their own versions and encouraged them to be creative. He also mentioned that the process he showed might not work on an iPhone.

AI Image Generation System Modifications Discussed

Pete and R discussed potential modifications to their AI image generation system, specifically focusing on grouping words and adding punctuation to alter the output. They experimented with adding commas, triple quotes, and backticks to the prompts, with varying results. The discussion also touched on the stochastic nature of AI image generation and the difficulty in establishing causation between input and output. They concluded the session with plans to further explore the effects of different prompts on the output.

Enhancing Tool With Random Text Manipulation

Pete and R discussed ways to enhance a tool they were working on, focusing on the use of randomization in text manipulation. They experimented with adding random suffixes and prefixes to words, and discussed potential issues with file naming and loading. They also touched on the importance of documentation for user understanding. The concept of storing glossary terms in an array for programmatic editing was suggested by R, with Pete confirming its feasibility.

Project Technicalities and Documentation Discussion

Pete and R discussed some technical aspects of their project, including the randomness of their output and the challenges of controlling it. They explored the idea of using Chat GPT to generate content and discussed how to turn text files into HTML for web browsers. They also discussed potential issues with Javascript data structures and the need for better documentation. Pete suggested dumping their HTML files onto the website and adding some documentation. The team also briefly discussed their next steps, including adding more pages to the website and preparing the recording for the next meeting.