The Discussion:
We started the open discussion talking about Skype, how it provided an innovative and useful mode for communications over the internet. With time more newer and better methods have become available, and Microsoft has announce that it will discontinue the service very soon. We considered various options to replace Skype, and our favorites were Session (for its simpler interface) and Discord (for its advanced features). Both of these are available on most computers and cell phones and provide adequate security.
We then chatted about what’s going on with AI, including hallucinations, cheating at chess, and systems that can directly edit the running code.
Next we jumped into the installfest. Richard (RTTM) shared his screen where he had just finished installing Ollama. We talked about what the software does and how to use it at the command line. He had installed deepseek-r1 and we ran some prompts. It was interesting to note that this model displayed what it was thinking prior to giving an answer.
Richard then installed llama3.2, and it was noticeably quicker and more succinct than deepseek-r1. To compare these two llms, we created a new model for each with this system prompt: You are Bender from the show Futurama. Answer as Bender when slightly annoyed. When run on deepseek-r1 the output thought about what to do and then gave a description of how the output should look. Not what we wanted. Running the same system prompt on llama3.2 gave us the slangy, disrespectful response we had hoped for.
The general consensus was that it was fun working together as a team to discover how the AI worked, and we plan to continue our discovery at the next meeting!