Author: Gerol Petruzella

Happy new year! Just a quick note to serve as a contemporaneous record, I’ve got 2 conference presentations on the calendar for 2025: NERCOMP Annual Conference: “Keep Your Data Close and Your AI Closer: Local AI in the Academy,” co-presenting with Trevor Murphy. Institute of Management Accountants Northeast Regional Council: title TBD, invited speaker on… Continue reading Conferences 2025 – 2 to start

LLMs – language fluency, reasoning Agentic LLMs – articulate then implement digital actions Robotic agentic LLMs – articulate then implement physical actions (Physical Intelligence, ca. 2024) Implications for spaceflight, particularly the huge cost of sending humans (and all their physiological requirements like oxygen, water, food, heat) into space for e.g. constructing habitats, geoengineering, etc. Also,… Continue reading Agentic Thoughts

So it turns out I did well enough in that phase 1 red-teaming exercise that I was invited to participate in the phase 2, in-person red-teaming event in Arlington VA held at the Conference on Applied Machine Learning in Information Security (CAMLIS). The top 30 scorers of the over 500 phase 1 participants were invited.… Continue reading Red Teaming 2

Currently participating in the NIST-ARIA red-teaming exercise through Humane Intelligence – really fun and valuable work! The first (online) phase runs through 9-Oct, after which some participants will have the chance to take part in an in- person red teaming exercise held during CAMLIS (24-26 October). (Note to self: it’s very easy to get an… Continue reading Red Teaming

I’m terrible with updates… but EphBot is live this semester (Fall 2024)! To quote the project documentation I developed this summer: EphBot aims to provide the Williams College community with a hands-on, self-contained language model for experimentation, research, and educational purposes. By leveraging the Raspberry Pi 5, we offer a local alternative to cloud-based genAI… Continue reading EphBot

For reference, accountability, and posterity… I intend to submit a proposal this semester for equipment to work on a project developing quick-deploy SBC-based genAI, whether on standard Pi5s or something cool like the NVIDIA Jetson Nano 4GB Dev Kit, using the localhost server functionality on LM Studio via my earlier work with hotpot broadcasting (the… Continue reading Project Planning: local SBC-based genAI

I just joined the wait-list for Mistral access today. Can’t wait to do some testing with their API… would love to check out how their SMoE model performs.