By: José Gregorio Argomedo, Advisory Technology Partner at RSM Chile

 

It's hard to believe that just a few months ago, we were introduced to ChatGPT, and even before its public release late in November 2022, a select group of companies had the privilege of early access to GPT-4 in mid-August 2022. These technologies are reshaping the way we interact with everything, from education and technological development to poetry; the use cases are limitless. 

 

Khan Academy, one of the privileged entities to have early access to GPT-4, plans to launch 'Khanmigo', an AI Assistant aiming to revolutionize education. The hope is that it could solve the 'Two Sigma Problem' (where individual tutoring raises a student's performance relative to a baseline class by two standard deviations) by providing a comprehensive AI tutor for each student and teacher [1].

 

I refer to this as the 'Language Revolution'. We are only beginning to grasp its power. To understand its origin, I recommend the pivotal paper 'Attention is all you need' [2]. In simple terms, transformers have revolutionized NLP, using attention mechanisms for understanding context. But this is just the beginning.

 

Research has shown that these models develop 'Emergent Abilities', abilities that were not explicitly programmed into the model but surfaced as the model grew in size and computing power[3]. For instance, next-token models developed the ability to perform math. It's fascinating to realize that these emergent abilities are unpredictable, unintentional, and still largely unknown - larger models are expected to continue unveiling new abilities.

 

Predicting the future is tricky. GPT-4 was made publicly available in mid-March, less than two months ago, and its impact on our work has already been significant. Greg Brockman, at the launch of GTP-4, showed how the model can understand the tax code and provide advice, also check Greg´s Ted Talk about the astonishing potential of GPT-4 [4]. To mention just a couple of other applications, Github's Copilot X, which assists with programming tasks, and BOX's AI assistant that helps manage documents. Microsoft and Google plan to integrate these technologies across all their services. 

 

Just imagine waking up in the morning to a message from your AI assistant - it's already read your emails, sorted them, and even created or modified meetings and tasks for your day. The potential for convenience and productivity is staggering. 

 

This is a call to action. To remain relevant, we need to understand these models - their abilities, limitations, biases, risks, and how they can impact our customers. This could be a growth opportunity if we harness its power effectively. While Chegg took a hit in its stock this week [5], Khan is planning its future with technologies like GPT at the heart of this new revolution.

 

I leave you with this quote from Bill Gates, “In my lifetime, I’ve seen two demonstrations of technology that struck me as revolutionary”. He was referring to the GUI (graphical user interface) and GPT[6].

 

[1] https://www.ted.com/talks/sal_khan_the_amazing_ai_super_tutor_for_students_and_teachers/c

[2] https://arxiv.org/pdf/1706.03762.pdf

[3] https://www.jasonwei.net/blog/emergence

[4] https://www.ted.com/talks/greg_brockman_the_inside_story_of_chatgpt_s_astonishing_potential/comments

[5] https://www.cnbc.com/2023/05/02/chegg-ceo-chatgpt-stock-plunge-was-extraordinarily-overblown.html

[6] https://www.gatesnotes.com/The-Age-of-AI-Has-Begun