388: Can AI be Controlled?, Okta's Most Popular Apps of 2022, Jassy's Amazon, Berkshire Drops TSMC, Onlyfans, and Detroit Pizza
"You know what they say about agency: Use it or lose it!"
Anything real begins with the fiction of what could be. Imagination is therefore the most potent force in the universe, and a skill you can get better at. It’s the one skill in life that benefits from ignoring what everyone else knows. -Kevin Kelly
👨🏻🍳👩🏼🍳🍕 My wife and I made Detroit-style pizza for the first time (I wrote about discovering this style of pizza back in edition #330).
Home-made dough, sauce, everything!
It smelled great while making it, and it was delicious to eat. I can think of ways to improve it for V2, but it was a successful experiment.
If you want to try it: We used Ethan Chlebowski’s recipe.
🗣️🗣️🗣️🗣️ Prediction time: Onlyfans creators will increasingly use large language models (LLMs) like ChatGPT.
I bet some startup is working right now on an unrestricted model that can be fine-tuned to mimic a messaging style based on past conversation logs.
These creators are largely in the parasocial relationship business, so having the ability to have “conversations” with hundreds or even thousands of customers at the same time will be a huge advantage to increase paid conversions and reduce churn.
See also this interview I did:
🏃♂️🗓️ I went running for the first time in a little over a month.
It feels good to get back in the saddle.
At first, I had valid reasons for not going — the past month had all kinds of strange weather in my area, from rain to -40°, uneven ice on sidewalks and roads, tons of snow — however over time it was starting to feel like I was losing the habit of running AND replacing it with the habit of finding excuses for not going.
This last point is important: You don’t just lose habits, you also create new anti-habits that fill the void. To get back, you have to not only recreate the original habit but also remove the new anti-habit.
🤖🤔 I often hear that LLMs are “just” predicting the next word.
There’s no “just” about it.
That’s like saying that human thinking is “just” predicting the next thought based on “memory retrieval and pattern recognition”.
I’m not saying there’s a one-to-one equivalence between human thought and a LLM deciding which token comes next in a sentence, but there’s nothing simple or obvious about answering someone’s question in a way that makes sense.
People who say something is “just X” are often short-circuiting deeper thought about what is actually taking place by providing a non-answer answer that superficially fills a curiosity void.
Don’t stop asking “why” and “how” too quickly...
🐲🧙♂️🧝🏻🧟♀️⚔️🛡️ Friend-of-the-show and Extra-Deluxe (💚💚💚💚💚 🥃) support Damien C. Tanner sent me something quite interesting.
Some friends played a game of Dungeons & Dragons with GPT-3 playing the role of the dungeon master (DM — the person who tells the story and controls all non-player characters and monsters):
It was a nice proof-of-concept, albeit on a small scale, but it demonstrated the potential there!
It’s something I had hypothesized about in the past, and my own D&D group has used GPT-3 and ChatGPT a few times to create some story elements to assist the DM (though not replace it).
I think it’ll really work when you can easily hook up one of these models to a voice-to-text and text-to-voice system (ie. an Amazon Echo) so that players can speak to it directly and hear the descriptions and answers read out loud (maybe even with an AI model that provides voice inflection and makes it sound more like a voice actor!).
Ideally, this AI model would be fine-tuned on all the rule books, the story modules and various expansion packs, and a whole bunch of transcripts of existing D&D games (from podcasts and Youtube?).
Now *that* would be interesting to see in action!
I suspect that engineering good meta prompts would also go a long way.
For example: “You are a great dungeon master in the role-playing game D&D, you make vivid descriptions, create great NPC dialogue with memorable flavor, and you have a kinetic way to describe fight sequences. You encourage emotional bonds between characters and create stakes through story for better payoffs…”
🏦 💰 Liberty Capital 💳 💴
🤖 From Bing to Sydney to Venom to HAL9000, oh boy 😬
Keep reading with a 7-day free trial