Ep 3. Our approach to AI is broken | George Morgan

Описание: In this episode we talk with ⁠George Morgan⁠ who is the founder and CEO of ⁠Symbolica AI⁠. Symbolica is a moonshot AI company. George believes the way that AI models are trained today is fundamentally broken and will not lead us to AGI. Symbolica is taking a completely different approach that will make models orders of magnitude faster, cheaper, and with additional reasoning capabilities. 

George is also a fascinating human being and this episode is packed with stories about Elon Musk, Andrej Kaparthy, the Sony QRIO robot, and much more. Of course, we also chat about our predictions on AGI and the future of AI. 

Hope you enjoy this conversation with George! 

* *
FULL SHOW NOTES
https://blog.infinity.ai/ep-3-our-approach...

* *
TIME STAMPS
02:30 The 20-year robot saga that took George to Japan 
10:55 What’s wrong with AI models today? 
16:40 What does it mean to represent things symbolically instead of statistically? 
20:20 Symbolica does not have labels, only data
25:10 The manifold hypothesis 
29:44 What is online learning? 
34:58 Symbolica’s moonshot goals 
38:04 What will be Symbolica’s ChatGPT moment? 
39:51 George’s path to Tesla (including dropping out of college and writing an email to Elon Musk!)
44:35 Why it’s stressful to work for Elon Musk 
47:21 Why George disagrees with Andrej Kaparthy 
51:35 The difference between being a founder and an employee (this is the #1 lesson George learned from Elon) 
55:50 George’s definition of AGI 
57:35 Common issues with LLMs: they mimic human behavior, but don’t let that fool you 
01:01:28 Lina’s different take on LLMs’ abilities 
1:04:02 George’s counter-argument 
1:09:09 Our discussion about a quote from “The Beginning of Infinity” (see below)


* *
QUOTE FROM "THE BEGINNING OF INFINITY" BOOK
“Even if chatbots did at some point start becoming much better at imitating humans (or at fooling humans), that would still not be a path to AI. Becoming better at pretending to think is not the same as coming closer to being able to think. [...] A non-AI program cannot fake AI. ”
— David Deutsch, “The Beginning of Infinity”

* *

LINKS TO RESOURCES MENTIONED IN THIS EPISODE:
Sony QRIO Robot: https://spectrum.ieee.org/qrio-the-robot-t...
Symbolica AI: https://www.symbolica.ai/
The Manifold Hypothesis: https://en.wikipedia.org/wiki/Manifold_hyp...
Tesla Self-Driving Team: https://www.tesla.com/AI
Elon Musk: https://twitter.com/elonmusk
Andrej Kaparthy: https://twitter.com/karpathy
Paper about “The Reversal Curse in LLMs — LLMs trained on “A is B” fail to learn “B is A”: https://arxiv.org/abs/2309.12288
Paper where GPT4 explains neurons in GPT2: https://openai.com/research/language-model...

* *
GEORGE'S ACCOUNTS
Twitter: ⁠https://twitter.com/vr4300⁠ 
LinkedIn: ⁠https://www.linkedin.com/in/georgemorgan2/⁠ 
Symbolica AI: ⁠https://twitter.com/symbolica_ai⁠    

LINA'S ACCOUNTS
Twitter: https://twitter.com/lina_colucci
LinkedIn: https://www.linkedin.com/in/lina-colucci/

* *
THE EDGE OF INFINITY: IMAGINING THE FUTURE OF AI
Will AI change the world? Everyone agrees that AI will change a lot about society, but no one knows exactly how. My name is Lina Colucci. I founded Infinity AI and got my PhD from MIT and Harvard. I talk to other AI founders, engineers, and researchers who are building the future of artificial intelligence. Together, we predict the ways AI will shape our world both in the near-term and as we project out to infinity.

Join us and let’s imagine the future of AI together.