Ethan Mollick on voice interaction with LLMs; another pseudo-acquisition; would you like to speak with an ai clone of me?; a robot factory worker; and a robot dentist
No doubt AI has benefits and more will develop, some quickly, some maybe more slowly. With that as my starting point, I've heard mention of a particular AI weakness only once. I'm not certain it's true and have no idea if it will remain true but it seems rather limiting for some current uses. I've heard that AI has no short-term memory. It has no memory of a conversation up to that point and no memory of a conversation with that person yesterday or last week. If true, this seems a huge gap in capability as well as in "intelligence."
Is this technically true? Is it the practical issue I perceive? Is it an issue likely to be solved anytime soon?
"I still make my bet on these sorts of specialized robots will be more practical than a generic humanoid robot."
I thought you previously said an AI robot needed to be able to do lots of tasks. I think the example was a kitchen robot? If that's accurate, the statements are not mutually exclusive but maybe you can explain further?
I'd probably wait a generation (which feels like 18 months at this point) before cloning yourself. The largest context windows are 1.5 million tokens, or about a million words, which is just enough for all of your books plus some chunk of blog posts, but if you got all your writing, all your class materials, every lecture you recorded, and every book.... Heck maybe even every email it would get just that bit closer to the real thing. A year of blog posts is probably a hundred fifty k tokens, and you have close to twenty years worth.
Sure. I'd like to discuss econ with clone you. What sort of feedback are you looking for? I will look at is a way to get feedback on some of my peculiar macro/Fed/inflation views.
I am curious, does the bot report to Arnold what was talked about, a transcript or something? I think I would want to know what clone me was telling people in my voice. (I also would want to know whether there were clones of me that I don’t know of.)
Thanks for the update! I am curious about how well it extrapolates what ASK would say about topics he hasn’t specifically discussed. Like if asked about Libertarian Topic B, which ASK has never addressed, does it sort of stitch something together based on what other people who voice similar opinions to ASK on topics X Y and Z that he has expressed? That could get pretty squirrelly when one has beliefs and opinions that do not map easily to common sets.
Please invite me to the clone.
2024 AI is a disaster; it does not make people more productive* & corporations are abandoning their 'AI strategies' [COUGH]
UPSHOT: The rubes went up a dead-end street
* "We see computers everywhere but in the productivity statistics" Robert Solow, 1987
Yes, please add me to your AI clone list. Thanks!
I would like to road test the Kling clone.
I would like to try the Kling clone.
No doubt AI has benefits and more will develop, some quickly, some maybe more slowly. With that as my starting point, I've heard mention of a particular AI weakness only once. I'm not certain it's true and have no idea if it will remain true but it seems rather limiting for some current uses. I've heard that AI has no short-term memory. It has no memory of a conversation up to that point and no memory of a conversation with that person yesterday or last week. If true, this seems a huge gap in capability as well as in "intelligence."
Is this technically true? Is it the practical issue I perceive? Is it an issue likely to be solved anytime soon?
"I still make my bet on these sorts of specialized robots will be more practical than a generic humanoid robot."
I thought you previously said an AI robot needed to be able to do lots of tasks. I think the example was a kitchen robot? If that's accurate, the statements are not mutually exclusive but maybe you can explain further?
By specialized I mean optimized for a particular environment, such as a kitchen or a dental office. Not necessarily shaped like a human.
Arnold you should at some point silently allow the clone to write your essays for a while to see if it fools us.
I'd probably wait a generation (which feels like 18 months at this point) before cloning yourself. The largest context windows are 1.5 million tokens, or about a million words, which is just enough for all of your books plus some chunk of blog posts, but if you got all your writing, all your class materials, every lecture you recorded, and every book.... Heck maybe even every email it would get just that bit closer to the real thing. A year of blog posts is probably a hundred fifty k tokens, and you have close to twenty years worth.
I would love to try your clone!
I would very much like to talk to you, if not in person, then in clone.
I would try out an Arnold Kling clone!
Yes! I was most impressed by the pauses for breath in the responses!
Sure. I'd like to discuss econ with clone you. What sort of feedback are you looking for? I will look at is a way to get feedback on some of my peculiar macro/Fed/inflation views.
You need my email? It's t[last name2000] at yahoo
I am curious, does the bot report to Arnold what was talked about, a transcript or something? I think I would want to know what clone me was telling people in my voice. (I also would want to know whether there were clones of me that I don’t know of.)
Your question raises interesting privacy concerns.
Thanks for the update! I am curious about how well it extrapolates what ASK would say about topics he hasn’t specifically discussed. Like if asked about Libertarian Topic B, which ASK has never addressed, does it sort of stitch something together based on what other people who voice similar opinions to ASK on topics X Y and Z that he has expressed? That could get pretty squirrelly when one has beliefs and opinions that do not map easily to common sets.
That's Cameron Crowe, before anyone phones in.
What operations does the calculator app do?