A chat on LLM, ChatGPT
An informal presentation by Anil Ananthaswamy in Bangalore.
Some elements of the talk:
- comparison of current AI model complexity with the complexity of the average human brain,
- MATH dataset, GPT3 failed in 2021, Google Minerva passed 50% in 2022
- What does “language” mean — debates
- Some inputs on “Theory of the mind”
Audience/author seem to be of similar opinion — concerns on pace of development, impact on epistemology.
Unfortunately, not much in terms of new perspectives on how humans might/should evolve, or whether there should be control on development of AI.
Perhaps, as in the last few pages of “Will I Be Mine” — something might emerge.
Across the world, governments voted to shut down labs that operated at the boundary of machine and human intelligence without having sufficient regulations in place. Though the reason cited was insufficient funding, insider sources knew that these labs were all involved in similar areas of research — that of improving Artificial Intelligence and melding it with Human intelligence, without any governing rules to manage conflicts.
But everyone knew, a storm was coming with or without their rules, and there was very little time.
