Weekend Listen: Anthropic Founder & CEO Dario Amodei
Not our podcast, but worth a listen if you spend time working on and/or investing in the Gen AI space (HERE).
Amodei said it is a 60/40, perhaps a 70/30 bet that future Gen AI models will operate at the level of a Nobel winner. If future models do not achieve that level, then Gen AI will be a bust (his words, not ours).
Time will tell if Anthropic, OpenAI and Google hit a wall in terms of the gains made by each model generation. To that end, the forthcoming OpenAI model - GPTNEXT - is reported to have 100x the computational volume of its predecessor model - GPT 4.
The question is: will the computational improvement as measured by Orders Of Magnitude (OOM), continue at what has been the pace of improvement over the past decade, or, will Gen AI model gains slow? There is not a Moore’s Law equivalent as it relates to the computational volume and processing power of Gen AI models. Amodei is upfront about this. He does not know for certain. Therefore, how can investors be so certain about prospective economic productivity gains derived from Gen AI?
Also, keep an eye out for Groq. Inference is everything (inference is the most expensive part of the Gen AI cost equation) and Groq’s LPUs are less expensive than Nvidia’s chip arrays.