The Discussion:
We warmed up by talking about some of our favorite tv shows, sci-fi movies and special effects.
Then the topic of rail guns came up. Not the cannons mounted on train cars during WWII, but guns that accelerate metal objects to incredible speeds. YouTube has a clip about a large unit that is no longer being developed, and Arcflash Labs has several models of hand-held units.
Gib then talked about how he created a video with AI generated characters using an AI storyline based a very short description. He showed us how to use the LTX Studio website to create, edit and reinvent AI videos. This service provides limited free usage to see it in action.
Carl talked about the importance of providing a context when using an AI prompt. For example, a prompt like “1966 Mustang” could produce anything from a car driving down the road on a sunny day to an off-road moon-lit silhouette. To get a consistent style when creating a set of images, consider adding a theme and/or a style in the prompts. For example “sleek futuristic nighttime image with neon highlights” could be used before “penguin driving car”, “penguin checking mailbox”, and “penguin meeting with friends”.
Pat talked about how it was difficult to find instructions for creating an LLM from scratch. So Carl, just for fun, gave this prompt to phi4: give me a step by step guide to create an LLM. no explanations, just the list of commands to run. This provided an 8 step guide complete with python code that could be copied and pasted.
Then Carl ran the same prompt on deepseek-r1, and it provided a list of instructions. However there were no commands provided. In this case, phi4 completely outshines deepseek.
Richard (RTTM) showed some of the artwork he has produced, including a proposed new logo for MDLUG. He talked about installing his own chatbot, and the suggestion was made the we do an AI installfest next time. Several of us will how we installed our chosen systems and answer any questions about how to replicate them on your own computers.