Discover more from Liberty’s Highlights
408: AI & Alzheimer's, Steve Jobs, Why LEGO Won, Buffett on TSMC, Nintendo, BloombergGPT, Amazon, Sam Hinkie, and The Godfather
"maybe intelligence is its own thing"
Do not confuse things that are hard with things that are valuable.
Many things in life are hard. Just because you are giving a great effort does not mean you are working toward a great result.
Make sure that mountain is worth climbing.
📘 🏋️♂️👦🏻🥋 I first mentioned reading the Warrior Kid books to my son back in Edition #243. We took some detours and read other things in between, but we came back to the series and recently finished the 3rd book.
My son loved it, and I think the life lessons in this book are just as powerful as they were in the first two.
Jocko writes clearly and knows how to build on what came before to keep it interesting — every book feels familiar but has enough new things to expand the concepts.
You can never be sure about causality, but my son has adopted many of the concepts from the books (we talk about “strong food” and “weak food”, he’s been doing Jiu Jitsu for a few months now, and has been doing push-ups and pull-ups since reading the first book).
But the most important changes are probably less visible.
It’s great to eat well and get in better shape, but the mindset keystones that the books promote are likely even more valuable. Humility, discipline, respect, resilience, integrity, courage, hard work, perseverance, keeping our ego in check, continual learning and growth, etc.
I wish I had read those books when I was a kid.
🧠🤔💭🤖💬 A few days ago, we visited my wife’s grandparents.
Her grandmother has Alzheimer's Disease (AD), which is at an advanced stage, so she doesn’t recognize us and forgets what we told her within about 10 minutes.
You can predict everything I may say about the disease itself — it’s horrible and job #1 is to cure it and prevent it — so I want to talk about something else that my visit there made me think of.
It’s obvious that AI assistants/companions could make her life better as a supplemental thing to the people who help her. In a perfect world, there would be people to help everyone suffering from AD 24/7. But in reality, there aren’t enough people to help and/or this level of care is simply too expensive for most of the families who are afflicted.
But AI could scale to meet this need, at least until we cure this disease. AI never gets tired of answering the same questions, can hold conversations, and doesn’t mind if things sometimes go in circles. It can play games (cards, board games, etc), and could even answer personal questions if it’s trained on a person’s personal history/family data, etc.
It seems like an obvious use case, and I hope many startups and companies are working on this.
💚 🥃 🐇 This is the free edition of Liberty’s Highlights with 16,500+ subscribers.
You can get 2 extra editions/week full of juicy stuff + access to the private Discord 🗣🗣🗣 community by becoming a paid supporter (it’s quick & easy).
Paid post since last week 🐇:
If you click the links above, you can see the intros for free and there’s a link to get a 7-day free trial of the paid version. You can only gain from this!
🏦 💰 Liberty Capital 💳 💴
🍏 New book of Steve Jobs speeches & letters (it’s FREE) 📕
The official Steve Jobs Archive has just released a book of speeches, interviews, and correspondence by Jobs.
It also contains a bunch of great photos of Jobs, a lot of them candid polaroids rather than the press photos that we’ve all seen a thousand times. Even if you don’t read the book, it’s worth flipping through for the photos.
The price is hard to beat: You can download it in various formats for free.
🇹🇼 Buffett on partial sale of TSMC stake 🐜
Buffett also commented on the sale of the bulk of Berkshire's stake in Taiwan Semiconductor Manufacturing Co. After buying more than $4 billion in shares in the world's biggest chipmaker between July and September 2022, it slashed its holdings by 85% to $617 million by the end of 2022.
Buffett said geopolitical tensions were "a consideration" in the divestment. He described TSMC as a well-managed company but added that Berkshire had better places to deploy its capital.
As my friend MBI (💎🐕) pointed out, this makes me wonder how Buffett feels about Apple’s geopolitical risk (both because of the supply chain in China, and dependence on TSMC) 🤔
💾 Non-programmer built an app entirely using ChatGPT 🤖
As everybody is experimenting and trying to find the edges of the capabilities of the current crop of generative AI models — personally, I don’t think we’re even close, and even if no new models were on the horizon we’d still find new things in the current ones for years — it’s interesting to watch experiments like this one:
Lucas Crespo decided to try to build an app without any programming experience just by asking ChatGPT to create the code and explain to him what he needs to do to stitch everything together, and to fix bugs and problems as they arise.
It’s not exactly easy. He had to install Xcode and do some debugging, but the fact that it’s possible at all is kind of mind-blowing and shows where things are going.
Why did LEGO win if it didn’t invent the… Lego block 🧱
Who here had heard of “Kiddiecraft”?
I certainly didn’t know this. It goes to show you that having a good idea isn’t enough.
🕹️ Nintendo’s Modern History: The Console Wars 🍄
I *loved* part 1 of Nintendo’s history by my friends at Acquired (💚 🥃), and so part 2 is self-recommending.
I’ve only had time to listen to the first 2 hours, but it’s very good. I like the story of how they created the Gameboy and where they got the name (hint: think about what Sony was selling a lot of at the time), the story of the fight against Sega was fascinating (love the counter-positioning), and I’m very curious about the story of the Switch juggernaut!
This isn’t just business history, it’s also great storytelling!
BloombergGPT’s 50-billion parameters finance-specific model 🏦🤖
As a financial data company, Bloomberg’s data analysts have collected and maintained financial language documents over the span of forty years. The team pulled from this extensive archive of financial data to create a comprehensive 363 billion token dataset consisting of English financial documents.
This data was augmented with a 345 billion token public dataset to create a large training corpus with over 700 billion tokens. Using a portion of this training corpus, the team trained a 50-billion parameter decoder-only causal language model.
Of course, they claim that their model outperforms “similarly-sized open models on financial NLP tasks by significant margins”.
It’ll be interesting to see if the company’s data truly differentiates it, or if general/foundational models can be fine-tuned with non-proprietary data and achieve competitive performance.
Friend-of-the-show and Extra-Deluxe supporter Byrne Hobart (💚💚💚💚💚 🥃) wrote some great stuff on BloombergGPT. Some highlights:
At their core, financial data companies are really data cleaning companies, whose economics are often driven by the fact that the window of time available to exploit some information is longer than the window of time needed to acquire it. They're all constantly opportunistically getting new data [...]
One of the use cases they highlight is a good example of how these kinds of models can create and perpetuate lock-in: BloombergGPT is, naturally, very good at converting natural language questions into queries in Bloomberg's own Bloomberg Query Language, which has historically been used in their Excel add-in and other APIs. That makes it much easier to build products that are continuously dependent on Bloomberg, and at least part of the company's recurring revenue comes from the fact that clients have models that rely on Bloomberg data even if there's a cheaper alternative.
🛒 Amazon starts charging for some UPS returns… 📭
Amazon has started charging fees to make some returns at UPS Stores, marking the latest effort by the e-commerce giant to cut down on costs associated with customers sending back items.
The year of efficiency continues!
🤖🏗️💵 Avi Goldfarb on the Economic Impact of AI (Podcast)
I enjoyed this discussion about some of the potential impacts of AI on the economy, and unlike a lot of the discourse lately, this one is very grounded and practical.
In some ways, it feels like a less ambitious vision of the AI future, but maybe it’ll turn out to be the more realistic one, who knows? We often under-estimate change, but once in a while, we also over-estimate it…
I’m not sure what I think about all the historical comparisons with farmers and steam-powered factories and such, though.
It’s a beautiful analogy, very elegant. 📈📉
But I can’t help but think that maybe intelligence is its own thing and that the effects will materialize in ways that are less predictable than increases in mechanical power or widget-output-productivity.
The bottom line is: I don’t know, but it’s interesting to think about ¯\_(ツ)_/¯
🗄️ From the archives: Sam Hinkie on Playing-Long Games and the Talent Power Law (Podcast)
It’s kind of a podcast-heavy Edition, but I don’t really mind. Audio or text or video… That’s just the format. What matters is the information, the ideas, the insights!
I recently saw that Patrick (☘️) had uploaded an episode that combined some of his fave interviews together, and I saw Hinkie’s name in the title. It made me want to go back to the original source and re-listen to Sam Hinkie’s interviews:
I loved these when I first heard them, and they’ve aged well. I’ve learned more from listening again.
I love the focus on depth, on building relationships with extraordinary people, on understanding that talent follows a power law, and getting good at identifying rare talent and empowering it.
Even the basketball stuff: I don’t care about 🏀, but I loved ‘The Last Dance’ because it’s about a lot more: the meta-game, the craft, self-improvement, team building, pushing past obstacles, etc.
It all sounds a bit cliché, but it’s real stuff.
🧪🔬 Liberty Labs 🧬 🔭
Observing 25 AI agents living in a virtual town 🏡 ⛹️♀️🏡🚶🏻♂️🏡🚶♀️
Here’s a new paper that I found quite interesting.
The experiment was to create a Sims-like small town and let loose 25 AI agents with individual motivations and memories. The kind of complex behavior that resulted was impressive:
For the purposes of this paper, we focus on the ability to create a small, interactive society of agents inspired by games such as The Sims.1 By connecting our architecture to the ChatGPT large language model, we manifest a small society of twenty five agents in a game environment. End users can observe and interact with these agents. If an end user or developer wanted the town to host an in-game Valentine’s Day party, for example, traditional game environments would require scripting tens of characters’ behavior manually. We demonstrate that, with generative agents, it is sufficient to simply tell one agent that she wants to throw a party. Despite many potential points of failure—the party planner must remember to tell other agents about the party, attendees must remember the invitation, those who remember must decide to actually show up, and other possible points of failure—agents in our environment succeed. They spread the word about the party and then show up, with one agent even asking another agent on a date to the party, all from this single user-generated seed suggestion
You can see a recorded version of the simulation here (by clicking on the characters at the bottom, you can jump from one to the other on the map to see what they’re up to).
🌀 ‘No hurricane has ever crossed the equator.’ 🌐
Thank you Coriolis force!
And if you’re wondering about that lone storm in South America, it’s Hurricane Catarina:
Hurricane Catarina, or Cyclone Catarina South Atlantic tropical cyclone, the only hurricane-strength storm on record in the South Atlantic Ocean. Catarina made landfall on South Brazil at peak intensity, with the equivalent of Category 2 hurricane-force sustained winds, on 28 March 2004. [...] Catarina was the first tropical cyclone to make landfall in Brazil since the beginning of reliable records; hence, the infrastructure and population were not specifically prepared for it, which led to severe damage
🔦 The Speed of Light Visualized ⭐️ → 🌎
Even when you know how it works, seeing it visualized is kind of 🤯
It kind of looks like the Earth and the Moon are playing a game of Pong
🎨 🎭 Liberty Studio 👩🎨 🎥
🐴 🎬 ‘The Offer’ + ‘The Godfather’ (2 + 2 = 5)
It’s a long film, so my wife and I have only had time to watch 2/3 of it, but I’m really enjoying the journey — and discovering so much in the film thanks to the extra context from the TV series on how it was created.
I knew it was a masterpiece, and I had seen it over a decade ago, but it’s even better than I remembered.
In fact, I like both The Offer and The Godfather so much that I’ve decided to create a project about them. Stay tuned!
🗓️ Medley of the most popular songs each month from January 1980 to December 2022 📻🎶
Watching the changes in visual styles in the video is as interesting as listening to the changes in the music!