[03:42] Elias starts trading bonds after studying math, econometrics, and computer science.
[04:17] From notation calculators to basic spreadsheets to nascent AI, Elias sees patterns in tool evolution.
[05:17] Elias moves to consulting, always involved in quantitative fields.
[06:20] The significant AI break throughs since 2016-17.
[07:12[ Why self-supervised learning was one critical advance.
[07:50] New architectures--enabling much larger models—were a second step, leading to generative artificial (GenAI) models.
[08:55] What the “language” of Large Language Models (LLMs) covers.
[10:00] After training ChatGPT by absorbing the internet, “hallucinations” need to be eliminated.
[11:06] “Red teaming” to eliminate hallucinations.
[12:11] The next refinement step is “reinforcement learning from human feedback”.
[13:00] The issue of “jail-breaking” models to circumvent “blocked” answers.
[14:32] Data embedding or fine-tuning: using private data to train GPT.
[16:02] Why did ChatGPT stop data accretion in 2021?
[16:30] The considerable cost of topology, training, and refining AI models.
[17:43] User input in ChatGPT serves to refine the model more so than to teach it.
[19:37] The Future of Jobs: Will generative AI lead to mass job losses? If so, when?
[21:37] Why the impact of GenAI will be delayed in some areas.
[23:00] GenAI is impacting certain areas faster—such as coding and customer service—generally enabling significant productivity gains.
[24:35] Career progression must adjust as corporate pyramids’ bases shrink.
[26:00] Knowledge management will change appreciably, with new jobs created and new tools and processes invented.
[29:14] Different professions and companies try to codify their “secret sauce”—what can GenAI take care of?
[30:30] What will remain? How people show empathy, interact, and give emotional support.
[32:05] Many existing articles about GenAI contain factual inaccuracies.
[33:19] Training to understand applied technologies is becoming much more important.
[34:40] In a time of exponential curves, doom predictions are imprudent and never verified.
[35:18] What Elias is most excited about—especially leveling up the playing field.
[36:30] Likely effects: huge productivity improvements depending on the country’s social contract and a reduction in work time.
[37:40] Elias explains why timelines relating to GenAI are difficult to circumscribe and more than five years is now considered “long-term”.
[38:50] How Elias anticipates the dynamics of change over time due to GenAI.
[39:39] Why the “truth function” matters.
[40:26] AI may be capable of a kind of informed creativity, as humans do.
[40:44] The beneficial mix of technology, regulation, and internal company rules and the emerging need for a Chief AI Ethics Officer role.
[44:01] Misinformation is a major concern for Elias.
[45:22] The possible negative impact of generative AI on kids.
[47:02] We need a definition of what it means to be “human” and “intelligent”—remembering the movie “Her”.
[48:06] Comments on the open letter written by Musk, Wozniak, Harari, and others.
[49:47] What Geoff Hinton has achieved and what he has to say about GenAI.
[51:33] Fellow Turing Award winner Yann LeCun has a very different opinion about the potential impact of GenAI.
[52:25] Discussion on GenAI is something that will change at a fast pace: Elias will be back!
[54:04] IMMEDIATE ACTION TIP: Leaders must drive the change—identifying what impact gen AI will have at their company and articulating the vision of what the changes will look like--for change processes, teams, and more. Leaders must make it real with a roadmap and commitment to new behaviors, new skills and making them stick.
[55:08] As at other critical juncture points when so much is changing, many companies will need to rethink what they are doing and how they are doing it.
“People are often confused by the word language. They think only speech or text, but actually everything is language—code is language, music is language under certain constraints, an image is language.”
“If you’re talking about scientific questions, under the assumption that science is “true”, it’s very easy to say “Yes, this is true”. But when you arrive at political or, tomorrow, ethical questions, who determines what is true?
“What will remain, especially for client-facing professions, at the highest intellectual level or a lower intellectual level, will be how you interact with your client, with your customer. How do you show empathy and real interest, and how do you offer him or her emotional support?
“We live in an era of exponential curves. Everything evolves so rapidly that it's very difficult to predict when, how, and what the time horizons are. I’ve read some things about what AI will do in the next five years that I’m ready to bet will not happen.”
“If you ask an AI something about Galileo you can check in books if the answer is correct. But if you try to do prospective science, if you try to say “Build me a molecule that has these characteristics” and it comes out with a molecule, you will need to test if this is a real molecule.”
“Now, it’s another of those occasions. So many companies will need to rethink “What are we really doing? How are we doing it?”