September 19, 2024 – FS Insider speaks with Dr. Alan D Thompson, one of the foremost researchers and analysts in the AI space, about OpenAI's latest o1 model released just last week. Alan explains how o1 represents a 'new paradigm' for AI performance and reasoning, breaking nearly every test and benchmark we've developed. This is a really big deal, he explains in today's show, including surprising results on tests for self-awareness, deception, and power seeking as well. We dive into all the important aspects and key details of this latest development so be sure to tune in and share this info out with others!
Dr. Alan D Thompson publishes one of the most widely followed AI newsletters called the Memo, which is sent out to nearly every major lab, government, and the majority of Fortune 500s. You can sign up for it at lifearchitect.ai/memo.
Mentioned in today's show:
- o1: Smarter than we think (2024) – Dr Alan D. Thompson – Life Architect
- Alan’s conservative countdown to AGI – Dr Alan D. Thompson – Life Architect
- Mapping IQ, MMLU, MMLU-Pro, and GPQA – Dr Alan D. Thompson – Life Architect
- A Framework to Evaluate the Risks of LLMs for Assisting CBRN Production Processes
- OpenAI’s new model is better at reasoning and, occasionally, deceiving - The Verge
- Chat GPT’s new O1 model escaped its environment to complete “impossible” hacking task
