GPT/LLM links

Andy Kessler on Khanmigo; Timothy Taylor on productivity effects; Ethan Mollick on LLMs as book readers; The Zvi on deep fakes

Andy Kessler (WSJ) writes,

Mr. Khan’s stated goals for Khan Academy are “personalization and mastery.” He notes that “high-performing, wealthier households have resources—time, know-how and money—to provide their children one-on-one tutoring to learn subjects and then use schools to prove what they know.” With his company’s new AI-infused tool, Khanmigo—sounds like con migo or “with me”—one-on-one teaching can scale to the masses.

The column as many Neal Stephenson references.

Timothy Taylor writes,

think about all the technological changes of the last two decades–heck, over the past two centuries. Surely, if technological advances and automation were likely to lead to mass unemployment, we would already have arrived at a world where only 10% or fewer of adults have jobs? But instead, needing many fewer workers for jobs like growing wheat, lighting streetlights, filling out accounting ledgers by hand, operating telephone switchboards, making a ton of steel, and so on and so on have opened the way for new occupations to arise. I see no compelling reason why this time and this technology should be different.

Most of his post reports on research that attempts to forecast the effect of the new models on productivity. My prediction is that these effects will take a long time to appear. A “baseline” forecast would say that, because the record of past technological changes is that they took a long time to diffuse.

In the case of LLMs, I think that they are at a stage where they are hard to use. That may sound strange, considering how conversational they are. But you can think of LLMs as likes early personal computers or the early Web, where they were more like novelties than useful tools. People need to build the “killer apps” on top of LLMs.

I don’t want to have to copy-paste the content from each of my substack subscriptions into a conversation in order to get summaries from a chatbot. Instead, I want somebody to write an app that enables me to list my substack subscriptions and then, at a push of a button, give me a summary of everything posted from all of them each day.

It’s this app development that will bring about productivity gains using chatbots. And that will take time, with a lot of trial and error.

Ethan Mollick writes,

AIs have, or at least have the appearance of having, an understanding of the context and meaning of a piece of text. This radically changes how we approach books as sources of information and reference - we can ask the AI to extract meaning for us, and get reasonable results. These changes are exciting in some cases (there are amazing chances for scholarship assisted by AI), but threatening in others (why read the book when you can just ask an AI to read it?).

I can imagine feeding an LLM a book and then saying “Take out all of the throat-clearing, the academic jargon, the unnecessary homage to other academics, the cringey pop-culture references, and give me what’s left.” This could save me a lot of time. And maybe eventually intellectuals will stop writing books with lots of throat-clearing, academic jargon, etc.

Zvi Mowshowitz writes,

The core problem is, how do we verify what is real and what is fake, once our traditional identifiers of realness, such as one’s voice or soon a realistic-looking photo or then video, stop working, as they can be easily faked? And the cost of generating and sending false info of all kinds drops to near zero? How do I know that you are you, and you know that I am me?

He goes on to argue that this is a solvable problem. But he probably thinks that it is equivalent to solving the problem of stabilizing the deck chairs on the Titanic.

Substacks referenced above:

@

AI #13: Potential Algorithmic Improvements

@

What happens when AI reads a book 🤖📖