In this conversation with author and NYTimes reporter Cade Metz, I discuss the people and the companies behind modern AI and how they have shaped AI's agenda
I really loved it when Cade caught how the current GenAI works especially the ones where current LLMs lacks Formal reasoning which is true at this time, we've made LLMs smarter for Solving Math and Coding/Logic Problems using techniques like Supervised Fine tuning and Reinforcement Learning Verifiability which take in a lot of Clean Data and definitely Compute, but to gain real reasoining capabilites compute storage and data must not be a limiting constraint, it must be able to reason within limited data and optimize compute to come down to solutions than only we can say we've achieved or are on the path to true emergent abilities within the sphere of AGI, When Cade caught this, It shows his fundamental understanding of How NNs and LLMs actually work in reality and why we're still a long road ahead to go. Things like World Models and Spatial Models Developed by other Pioneers of AI Like Dr. Yan LeCunn and Dr. Fei Fei Li are currenlty exploring will make AGI more and more real and help the goal to have a smooth integartion with Humans, this can even be a great area of research in the Domain of HCI maybe.
The chemistry is better when there is an audience even though we do not see their faces.
(at 44:00) It concerns me that AI is replacing the developers who have shallow knowledge. The people (mainly students) I meet in that layer are not motivated to fill the knowledge gap. Will the system deteriorate over a period?
You observed on non-text inputs. Indian weapon systems used in Operation Sindoor (e.g., Akashtir) have used AI admiringly sans text input. I understand that there are considerable opportunities to use small data models in AI.
The transcriptions are excellent. Seem to be automatically edited to remove repeated words and grunts. Only a very few errors.
As you indicate, shallow knowledge is easily replaced by AI.
Regarding motivation: I was chatting with an educator recently about whether/how AI can help in education. His comment stayed with me: AI can help people learn but it cannot solve the motivation problem.
Dear Kartik Ji,watched entire1.05.13 "CI-02”—a meeting of Two Great(you & Mr.Cade Metz) ‘ai’ Minds. Learned in 1 ep.what it will take Reading many books.“saraswatirnihitadhi vaci" !! _परमेश्वरीं भगवतीं शारदा:_ resides in Sri Kartik Hosanagar's speech....!! WHARTON & your Optimism is a Contribution to society as AI is evolving at breakneck speed, transforming how value is created and delivered. Inaction is now a liability. Fifty years from now,when the AI history will be written,Wharton & yr bold and fast initiatives to promote ai,will also be written.
I really loved it when Cade caught how the current GenAI works especially the ones where current LLMs lacks Formal reasoning which is true at this time, we've made LLMs smarter for Solving Math and Coding/Logic Problems using techniques like Supervised Fine tuning and Reinforcement Learning Verifiability which take in a lot of Clean Data and definitely Compute, but to gain real reasoining capabilites compute storage and data must not be a limiting constraint, it must be able to reason within limited data and optimize compute to come down to solutions than only we can say we've achieved or are on the path to true emergent abilities within the sphere of AGI, When Cade caught this, It shows his fundamental understanding of How NNs and LLMs actually work in reality and why we're still a long road ahead to go. Things like World Models and Spatial Models Developed by other Pioneers of AI Like Dr. Yan LeCunn and Dr. Fei Fei Li are currenlty exploring will make AGI more and more real and help the goal to have a smooth integartion with Humans, this can even be a great area of research in the Domain of HCI maybe.
Trust DeepSeek will enter the discourse, albeit recent history of AI.
Your closing remark (optimistic) however cautious also highlighted the immense reskilling task ahead.
It is not just the lower end developers, most actors in the productive economy will need a AI upskilling.
An excellent interview/discussion, Kartik.
The chemistry is better when there is an audience even though we do not see their faces.
(at 44:00) It concerns me that AI is replacing the developers who have shallow knowledge. The people (mainly students) I meet in that layer are not motivated to fill the knowledge gap. Will the system deteriorate over a period?
You observed on non-text inputs. Indian weapon systems used in Operation Sindoor (e.g., Akashtir) have used AI admiringly sans text input. I understand that there are considerable opportunities to use small data models in AI.
The transcriptions are excellent. Seem to be automatically edited to remove repeated words and grunts. Only a very few errors.
My best wishes for the future episodes.
As you indicate, shallow knowledge is easily replaced by AI.
Regarding motivation: I was chatting with an educator recently about whether/how AI can help in education. His comment stayed with me: AI can help people learn but it cannot solve the motivation problem.
Dear Kartik Ji,watched entire1.05.13 "CI-02”—a meeting of Two Great(you & Mr.Cade Metz) ‘ai’ Minds. Learned in 1 ep.what it will take Reading many books.“saraswatirnihitadhi vaci" !! _परमेश्वरीं भगवतीं शारदा:_ resides in Sri Kartik Hosanagar's speech....!! WHARTON & your Optimism is a Contribution to society as AI is evolving at breakneck speed, transforming how value is created and delivered. Inaction is now a liability. Fifty years from now,when the AI history will be written,Wharton & yr bold and fast initiatives to promote ai,will also be written.
Thanks Sheo ratan ji.