326: Fintwit's Paradox, Thoughts on AI-Generated Art & Midjourney, Tesla's Lithium Refinery, China vs US, Cyborg Programmers, and U.S. Energy
"an agent representing the entire programming knowledge of the human race"
The idea that something can be likely and not happen, or unlikely and still happen, is one of the world’s most important tricks.
🚌 🎓 🧑🏼🏫 Follow-up to the link between learnings & fun from edition #325: Reader and supporter (💚 🥃) David Golberg shared his story with me, and with permission, I’m sharing it with you:
Re: learning, my daughter has been miserable in school. Struggling, calling herself dumb etc. We switched schools after academic testing to an amazing school the specializes in kids like my daughter. And after 3 weeks it’s a night and day difference. We no longer battle to do homework. She is happy, working hard. Just incredible what a positive approach does, by teachers trained to teach your kids in a personalized manner vs one size fits all. The Shelton School is a pretty incredible place. People from all over the country commute or move to Dallas to send their kids with learning differences to the school.
That’s so good to hear!
We need to learn from success stories, and have more variety of approaches available to better match kids, rather than the monolith that often feels more designed for getting kids out of parents’ way during the 9-5 workday rather than education.
🤖 🎼 How Long until we have a DALL-E or Midjourney for music? 🤔
I’ve thought about this every time I’ve heard music recently… There’s nothing about it that makes it inherently impossible. There may be a smaller training set than for images, and it may be harder to clearly disentangle and label various aspects, but hard doesn’t mean unattainable.
I know there’s already been AI-generated music (especially electronic music, for obvious reasons), but I’ve not seen anything as generalized and as malleable as the prompt-based visual AI art models… but I’m sure it’s coming.
Describe the genre, the mood, the tempo, the production style, lush or sparse arrangements?, what type of singing, the kind of lyrics (abstract poetry? Storytelling? Party vibe? Depressing?) and bam, the AI will spawn you a song.
A cross of Pink Floyd, Tool, Nick Cave, Bob Dylan, and Leonard Cohen
🛀 I just realized I haven’t heard the words “OpenSea”, “Cryptopunks”, or “Bored Apes” in weeks… I haven’t heard much about Waymo recently either, now that I think of it…
🇺🇦 Anne Applebaum — who was recently in Ukraine — wrote a good piece about the state of Russia’s war and what may lay ahead for both Ukraine and Russia. I recommend it.
💚 🥃 I appreciate that you are reading! I know everybody is busy and has countless options of things to spend their time on, so I take it very seriously that you decided to allocate slices of your life to my stuff.
The deal is, I can’t do it without you. Like you, I have bills to pay and mouths to feed, so it’s only win-win if we both get something of value. Thank you for your support!
Liberty’s Highlights is reader-supported. To support my work, consider becoming a paid supporter. I CAN’T DO IT WITHOUT YOU. Thanks! 🧙♂️
A Word From Our Sponsor: 📈 Revealera 📊
Revealera provides data and insights for investors into hiring trends for 3,500+ public/private companies + technology popularity trends for 500+ SaaS/Cloud Products.
We give investors insights into:
Job Openings trends: Insights into a company’s growth prospects.
Technology Popularity Trends: Insights into how widely products like Datadog, AWS, Splunk, etc, are gaining adoption.
Vendor Sign-ups (Currently Alpha) tracks the # of companies, as well as the specific companies, that have signed up for SaaS products such as Zoom in near real-time.
Visit Revealera.com for a ✨free✨ trial/demo.
🏦 💰 Liberty Capital 💳 💴
🐦 Fintwit’s Paradox ☯️ 🙃
Well, this isn’t just about Fintwit, but that was the shorter headline…
There’s a phenomenon I’ve seen on various investment forums. It goes something like this:
People complain that all everybody ever talks about are big, well-known companies (FANG, Berkshire, whatever). They say they want original ideas, under-the-radar companies, small caps with more growth in front of them than behind, cheap stocks with hair on them, etc.
When someone brings up something that fits this description, 19 times out of 20, it’s crickets 🦗, tumbleweed, and these tweets/threads sink without much of any reaction. 🌵
If you look at the archive of almost any investment forum, you’ll see that every thread with hundreds and hundreds of replies is a big well-known company, and at the very bottom with no or few replies are the “original, obscure ideas”.
Of course, it’s kinda circular to say that better-known companies are more discussed, but it’s still the case that the incentives in the system mean that these communities get very little of what they claim to want because they don’t reward those who go to the trouble of actually doing it. (there are exceptions like Xpel or whatever, but not that frequently…)
💡 Putting New Ideas Out vs Criticism 💣
While I’m on the topic, there’s a similar paradox when it comes to putting new ideas out there vs criticism.
You often hear in investing communities that people wish more people would discuss companies, stock ideas, etc, rather than endless discussion of news, macro, and memes.
But when people do, they often get so much crap — over and above healthy pushback and disagreement over ideas, I mean — that it discourages many thoughtful investors from sticking their necks out with ideas and opinions on stocks and businesses.
Much safer to complain about JPOW or whatever.
It may be because there’s so much less friction in criticism than sharing a well-formed and well-researched idea (it takes 5 seconds vs possible weeks and months of research), and because negativity and cynicism always sound smarter than staking money and reputation betting on some unknown future outcome.
So again, the rewards and incentives kind of discourage most people from sharing thoughtful ideas and analysis, which is why the few that do stand out so much (good for them, I guess!).
Shaping the conversation 💝
I want to put these ideas out because I think this is an emergent phenomenon that people rarely think about.
Nobody as an individual has the goal of doing this, they just unthinkingly do their thing, and we collectively end up pointed in that direction.
I figure that if people think about the situation a bit more explicitly, maybe they can act more deliberately and as a community, we can steer things in the direction that people prefer, even if only a little. That would be win-win…
Tesla in early stage of building its own lithium refinery in Texas 🔋🔋🔋🔋🔋
Tesla’s vertical integration may be about to take another big step forward:
Tesla said the plant would be focused on the development of “battery-grade lithium hydroxide” and be “the first of its kind in North America,” according to a newly released letter to the Texas Comptroller’s Office.
The company would process “raw ore material into a usable state for battery production” then ship the lithium hydroxide to various Tesla battery manufacturing factories, the application said.
Part of this move is no doubt to be less dependent on faraway supply chains (*cough* China *cough*), but also to try to secure large volumes at a time when everybody is bidding up the price of lithium (up 120% this year).
If Tesla’s application is approved, construction could begin in the fourth quarter of 2022, the company said. The project will reach “commercial operations” by the fourth quarter of 2024.
It’s still early in the process, and there are no construction contracts or permits yet. Something to keep an eye on — I wonder if this would open the floodgates and other automakers would start vertically integrating their battery supply chain more after this 🤔 TSLA 0.00
🇺🇸🏇 🇨🇳🏇China’s Slowdown, Bad Policies, and Demographics Mean it May Have a Hard Time Catching Up to the U.S. 🐢 🐇
China’s demographics, which I wrote about (with graphs) in edition #324, will be a huge headwind for the country. Add to that their other challenges, and the hare may actually have a hard time catching up:
Until recently, many economists assumed China’s gross domestic product measured in U.S. dollars would surpass that of the U.S. by the end of the decade, capping what many consider to be the most extraordinary economic ascent ever.
But the outlook for China’s economy has darkened this year, as Beijing-led policies—including its zero tolerance for Covid-19 and efforts to rein in real-estate speculation—have sapped growth. As economists pare back their forecasts for 2022, they have become more worried about China’s longer term prospects, with unfavorable demographics and high debt levels potentially weighing on any rebound.
Of course, if there’s one thing that is hard to predict, it’s the future, so who knows what could happen to the economies of both countries — positive and negative — in the coming years.
🧪🔬 Liberty Labs 🧬 🔭
🤓🤝🤖 Cyborg Programmers
Earth has too few people who know how to code vs the demand for software, but AI tools like GitHub Copilot and now Replit GhostWriter are going to make coders a lot more productive, and help smooth out the learning curve for new programmers.
AI Mode sports an ML-powered pair programmer that completes your code in realtime, tools to generate, transform, and explain code, and an in-editor search utility that lets you find and import open-source code without leaving your editor (think Stackoverflow in your editor).
AI Mode is like Multiplayer in that you collaborate in real-time with someone else. However, in this case, you're not coding with a person; instead, it's an agent representing the entire programming knowledge of the human race.
Check out the demos by Replit. Really impressive stuff, even on mobile!
via friend-of-the-show Fahd Ananta (🪳)
🏭 ⚡️ U.S. Energy Capacity Factors 🔌 💡
For those not familiar with the terminology, here’s what the Department of Energy says about ‘capacity factor’:
It basically measures how often a plant is running at maximum power. A plant with a capacity factor of 100% means it’s producing power all of the time.
Nuclear has the highest capacity factor of any other energy source—producing reliable, carbon-free power more than 92% of the time in 2021. That’s nearly twice as reliable as a coal (49.3%) or natural gas (54.4%) plant and almost 3 times more often than wind (34.6%) and solar (24.6%) plants.
This is important because depending on the source, there can be a pretty big difference between the nameplate megawatt number that makes the headlines and what you’re actually getting.
For example, a 1GW wind farm will produce around 300-350MW on average, because the capacity factor of wind is around 30-35% (higher in some places, lower in others). Nuclear in the US is around 90%, so a nuclear power plant with the same rating of 1GW will produce around 900MW on average.
Our friend-of-the-show at Enersection created very neat graphs showing capacity factors for various U.S. power sources over time.
You can see large versions of the graphs above on their page. Make sure to note that the Y axis starts at zero for everything except nuclear, which starts at 70%. If it was zero-based, it would mostly look like a straight line near the top of the graph.
Interesting to note how coal has been declining, from a very high base in the 70%-range to the 40s in recent years. It’s good to remember that even without shutting down coal plants, their coal consumption is going down over time as other cleaner sources take over.
🎨 🎭 Liberty Studio 👩🎨 🎥
Thoughts on AI-Generated Art & Midjourney 🖼 🎨🤖
I can’t stop thinking about Midjourney.
At the moment, I’m finding it more interesting than DALL-E 2 and Stable Diffusion (which I’ve recently installed on my Mac Studio) because it has an artistic quality that resonates with me.
The images above are just random ones that I saw in the firehose of people generating and upscaling things and liked enough to save. They aren’t even top images picked by the community (you can see the official showcase here — make sure you click “top” at the top right).
First, it blows my mind that a piece of software can generate these images from a text prompt in the first place. Many have such attention to detail when it comes to composition, use of colors, use of light and shadow, etc. Amazing that these somewhat abstract and subjective elements could be extracted from the training material and transposed to new compositions.
And it’s not like Midjourney is a one-trick pony that can only do digital painting or whatever. Some images are photo-realistic, some look like detailed 3D rendering with ray-tracing and depth-of-field, others look like oil paintings, some look like photos of marble sculptures, some are abstract patterns, etc.
What I can’t wrap my head around is how it can make something that looks like a 3D rendering without actually modeling things in 3D and rendering a scene.
It’s like it can skip to the last step, yet the lighting, shadows, and reflections don’t look all messed up and inconsistent! (sometimes they do, there’s plenty of weirdness and faces that are stuck in the uncanny valley, but the success rate is surprisingly high considering the difficulty, and rapidly getting better!). Look at this:
In fact, that’s the other thing that surprises me most: casually browsing other people’s images, I can’t believe how many of them are in the top decile of art that I’ve ever seen!
Artistic Super-Stimulus 🍫 🍪
In the same way that a chocolate bar or bag of Doritos are super-stimulus compared to what we’ve evolved eating, I’m starting to feel like AI art is honing in on artistic super-stimulus, being able to learn from a bunch of art that is generally regarded as being top-level, and creating more of it in seconds through trillions of calculations on a GPU.
There are “prompt engineers” who are excellent at creating requests that are dozens and dozens of words long, with lots of adjectives, to get the best results. Here’s an example that I saved:
horrifying vintage sci-fi art, retro::12 semi-dark 2D matte gouache illustration on black paper::9 gloomy dimensions of melancholy and big sad::11 detailed::10 luminous::5.9 intricate and complex::6.9 reflective chrome and bronze::8 balanced composition::7 atmospheric hyper-doom::12 --ar 2:3 --s 5000 --q 2
Most people start with a few words, but over time, you learn to “paint with words” and get better and more precise results using more specific prompts. It’s a brand new skill!
You can change the aspect ratio of images (make them square, vertical, horizontal, widescreen, etc), ask the AI to stylize things more or less, inject chaos into the process, or even ask for more or less compute time to be spent on the image.
And if you like something, you can have the AI re-compute it at a higher resolution (upscaling isn’t just making the image better, it actually adds real detail!).
As for myself, I’m a total newbie, but here’s one of the first images that I generated. Of course, it had to be Al Swearengen from Deadwood, in his saloon 🤠
This was the prompt I used:
Al Swearengen from Deadwood, HBO, standing behind bar of old western saloon, candlelight, oil painting, warm colors, brown, orange, brush strokes, wide shot