Next Gen
A lightning round (and return) of Prof. Mollick's insights into Artificial Intelligence
I want to return to Wharton Business School's Prof. Ethan Mollick to convey some of his lightning Q&A about Artificial Intelligence: 1. Can we detect AI writing / AI-generated images? (Not really.) 2. Who knows how to best use AI to help me with my work? (You do.) 3. Our company won’t let us use AI because we don’t want our data stolen, is that right? (It’s less a problem than you think.) 4. What’s the deal with copyright and AI? (It’s in flux, but large AI companies are eager to indemnify users – e.g, Google, Microsoft, and Adobe) 5. Aren’t AIs like GPT-4 getting worse with time? Or will halt as the internet fills with AI data? Or as it runs out of data to train on? (Nope.) 6. How good does AI get? (It’s unclear, but there is definitely more coming.)
On the last question “how good does AI get?”, these views last week caught my eye:
Andrej Karpath on LLMs as Operating System:
“With many pieces dropping recently, a more complete picture is emerging of LLMs not as a chatbot, but the kernel process of a new Operating System. … looking at LLMs as chatbots is the same as looking at early computers as calculators. We're seeing an emergence of a whole new computing paradigm, and it is very early.”
Alberto Romero elaborated:
“What this means, basically, is that chatbots are not the correct way to look at systems like GPT-4, or better, the future versions of these systems. A program that can talk to you as your friend or assistant or perhaps your partner or teacher is a tiny part of the puzzle.”
Rohit Krishnan gathered early examples of the “dream AI Hardware“:
And then … ChatGPT gave me a really answer insight when I asked “What would an operating system experience be if powered by ChatGPT?” (Just a little cut down for brevity.)
An operating system (OS) powered by ChatGPT would represent a significant shift from traditional GUI-based or command-line interfaces. The core idea would be to make the OS more conversational, intuitive, and context-aware. Users could interact with the OS using natural language. For example, instead of navigating through menus to find a setting, one could simply ask, “Change my wallpaper to a beach scene.”
The OS could understand the context of a user's actions. For instance, if you're working on a document and say, “Save this and send it to John,” the system would know which document you're referring to and who John is. Instead of manually searching for files, you could ask, “Where is the presentation I worked on last week?”
Technical Challenges: (1) Resource Utilization: Natural language processing is computationally intensive. The OS would need to be optimized to run these operations efficiently. (2) Misunderstandings and errors in natural language are common. The system would need robust error-handling mechanisms, possibly involving a fallback to traditional GUI elements for clarity and precision. (3) Security: Voice or text-based commands could be susceptible to spoofing or unauthorized access, requiring additional layers of authentication. Data should be encrypted, and the user should have control over what data is stored and how it's used.