Joseph Fluckiger's Blog

Imagine talking to someone who forgets half of what you said a minute ago. Frustrating, right? This is the challenge many artificial intelligence models face today. Artificial Intelligence and LLMs have made significant advancements in 2023, but one significant limitation is the Context window size limits. Context window is analogous to short term memory. When you ask a question to an LLM, it can only remember so much about the context of your question. A recent research paper introduces memGPT, which aims to provide AI with unlimited short term memory. The Challenge of Limited Memory in AI AI models have a fixed context window that restricts the amount of information they can process. While improvements have been made, these models still have limitations. MemGPT proposes a system inspired by the memory hierarchy in traditional operating systems. Introducing memGPT: Towards LLMs as Operating Systems MemGPT effectively manages its own memory through function calls, allowing the AI model...

I've used Excel for over 20 years. I've used Excel for everything from personal finance, to data center capacity planning. I even wrote a reporting program based off of Excel which used Excel formulas to render data and images! The application was TraceFinder and was used in the 2016 Olympics in which our ThermoFisher Analytical Chemestry instruments and reporting software were use to ensure that Usain Bolt and Michael Phelps and the rest of the Olympic athletes were drug free during the competition. This was an awesome application made possible by the power of Excel. But Excel formulas are a pain to write, and nowhere near as powerful as a fully turing complete language like Python which is the workhorse of DataScience.  If you've used Microsoft Excel, you're among the 750 million people across the globe who rely on this powerhouse productivity app. The recent integration of Python into Excel, announced by Microsoft, is an exciting development that is poised to elevate...