Discover more from Liberty’s Highlights
362: Stratechery, ChatGPT, Twitter, TSMC, Salesforce, AI Training Costs, Nvidia, Malaria, and Hendrix
"Sometimes you have to improvise!"
If you never fail, you’re only trying things that are too easy and playing far below your level... If you can’t remember any time in the last six months when you failed, you aren’t trying to do difficult enough things. -Eliezer Yudkowsky
📸 🤓 I am very glad that I took the time to develop basic photography skills.
Every time I look at photos of my kids, my wife, nature, or anything else, I’m thankful that I learned a few techniques and ways of thinking visually that help things look more interesting (at least to me).
The exact same subject in the exact same location with the exact same light can look bland and boring *or* interesting and rich depending on a few simple changes.
Sometimes just moving a few feet to the right or left can make all the difference between flat light or rich shadows. Going low near the ground or up in the air, putting the subject off-center, changing zoom level, etc
Like the rest of life, a few decisions can change everything.
But the nice thing is that once you’ve learned how to think about that stuff, you will continue to reap the rewards from those skills for the rest of your life. It doesn’t take more effort, it makes it more fun!
If you’d like to level up your photos a bit, I recommend this guide:
The content isn’t necessarily that different from other guides, but I love how images are used to illustrate the points. They’re worth kilo-words.
⚡️🔌🚫 No electricity at my house today. I’m writing this from a friend’s kitchen table. Sometimes you have to improvise!
🍎👴🏻🧓🏻 Wanna feel old?
The original Macintosh was released closer to the start of World War II than to today.
💻🤖💰 How much is Jensen Huang smiling when he thinks of millions of people creating images with generative AI and typing stuff in ChatGPT?
The amount of compute demand coming from consumers (as opposed to enterprises) must be going vertical. NVDA 0.00%↑
🐦💡 Here’s how I think Twitter should implement longer tweets to replace threads, which are a bad experience for both readers and writers.
I’m picturing that long tweets would all have an intro that is regular tweet length, and you’d need to click to expand and see the rest.
If you’re not interested by the intro tweet, you can just move on, but if you do want to read the longer form, you get a much smoother format than a series of 25 tweets.
The input screen could look something like this:
And the tweets would look the same except for an “Expand” button that people would over time get used to recognizing as the thing to click to see more. No more “🧵👇”.
Threads would still exist, but they’d probably be more for discussions or spontaneously thinking-out-loud ideas. For anything more complex and polished, this format would be win-win for writers and readers..
💚 🥃 🐇 This is the free edition of Liberty’s Highlights with 13,500+ 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 posts since last week 🐇:
This one has a lot of Amazon: My highlights from a long interview that CEO Andy Jassy gave and from the long AWS re:Invent keynote where they introduced lots of new products and services.
I go in-depth into Crowdstrike’s Q3 earnings call and a few recent Cloudflare presentation transcripts. Pipelines and electricity grids, etc.
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. What have you got to lose?
🧩 A Word From Our Sponsor: Heyday 🧠
Do you have 100+ browser tabs open right now? 😬
Give your memory a boost with Heyday, the research tool that automatically saves the content you view, and resurfaces it within your existing workflows. 👩💻
It’s like cheat codes for your memory. 😲💡
🏦 💰 Liberty Capital 💳 💴
🔥 10 years of Stratechery with Ben Thompson on Acquired Podcast ⏳
🤖 Friction: When quantitative differences become qualitative, ChatGPT Edition 📑
ChatGPT, the main character on the internet this week, is a good example of an important fact:
Friction matters and quantitative differences in friction become qualitative past certain thresholds.
Make something harder to use, and at some point it becomes useless to most people even if the capabilities remain the same.
Make something easier and more intuitive to use, and the reverse can be true.
While OpenAI did upgrade its model recently (the so-called GPT-3.5), most of it is still running what was originally trained in late 2020. What unlocked its capabilities to orders of magnitude more people were a new chat interface, some tweaks using reinforcement learning from human feedback, and the ability to keep some context from previous prompts (I think it just re-injects them into the new prompt).
Anyway, my point is not to rehash what has been written about how impressive ChatGPT is — Ben Thompson 💚 🥃 🎩 has a good piece on how we may make lemonade from the LLM unreliability’s lemons 🍋 by training humans to be more skeptical and better fact-checkers and Ethan Mollick had the AI write a university course syllabus for him — but to get you thinking about where else changes in friction can have similarly non-linear effects. 🤔
TSMC to increase investment in Arizona from $12bn to $40bn 🐜 💰💰💰💰💰💰💰💰💰💰💰💰💰💰
The Taiwan Semiconductor Manufacturing Company (TSMC) is set to announce it will increase its investment in Arizona from $12 billion to $40 billion when President Biden visits the chip manufacturer on Tuesday.
It will be the largest foreign investment in Arizona history and one of the largest in U.S. history.
That’s big. It’s the Supersize-Me of fabs.
The TSMC plants will produce 600,000 wafers per year when fully operational, which is enough to meet U.S. annual demand, according to the National Economic Council.
The U.S. plants will be a small fraction of TSMC’s total capacity, which produced 12 million wafers in 2020.
The main question is where will they find all the skilled labor required to operate these fabs. I understand that TSMC employees from Taiwan will work there, train people, etc.
But how long does it take to train super-specialized engineers in this type of black magic? Is there even enough supply in the US pipeline? How about the supply chain for this much semiconductor equipment? Also, fabs use a lot of water, which Arizona isn’t exactly known for 🌵 ¯\_(ツ)_/¯
Now may be a good time for the US to more fully take advantage of its immigration privileges and make it easier for the talent from the rest of the world to move there.
TSMC will also announce it will build a second facility in Arizona that will produce 3-nanometer chips by 2026.
‘Slack CEO Stewart Butterfield is leaving Salesforce two years after deal was announced’
This is a good reminder that most founders and entrepreneurs just can't handle being inside a big company unless they're in charge.
Most of them endure it until they vest their options and then leave.
I’m sure many are genuinely convinced that they’ll stick around when they first sell their company to some behemoth, but after a honeymoon period, the soul-crushing bureaucracy + meetings + slower pace + lack of focus at most big companies must break the illusion.
“I’m not going to do anything entrepreneurial,” Butterfield wrote in the Slack message. “As I said in my announcement to Slack team, these days my fantasies are about gardening. As hackneyed as it might sound, I really am going to spend more time with my family (as well as work on some personal projects, focus on health and generally put time into those things which [are] harder to do when one is leading a large organization).”
Does anyone really believe that?
Look at his Wikipedia page. The guy has founded or co-founded 3-4 companies (including Flickr! That brings me back to the heyday of Web 2.0).
I’m not saying he’s lying, he probably wants to slow down and smell the roses.
But chances are, after some time to recharge his batteries, he’ll get the itch again. Most people he knows are probably working in tech startups and building products. That’s what they’ll talk about over matcha lattes or whatever they drink, and at some point, something will make him go “hmm, I wonder why nobody is doing XYZ.. I wonder if…” 💡
Wouldn't it be amazing if Satya Nadella somehow could convince him to join Microsoft and be in charge of Teams? Not that it would ever happen... but what a plot twist this would be! 🤯 CRM 0.00%↑
How much does it cost to train large AI models? 🤖💸
Timothy Prickett Morgan found pricing data from Cerebras Systems and created this table showing his estimate of the cost to train large transformer models of various sizes, from 1.3 billion parameters to the full 175bn GPT-3:
🧪🔬 Liberty Labs 🧬 🔭
🩸🦟 Monoclonal Antibody vs Malaria 👩🔬
Promising study on the use of a monoclonal antibody to prevent malaria infections:
CIS43LS is a monoclonal antibody that was shown to protect against controlled Plasmodium falciparum infection in a phase 1 clinical trial. [...]
We conducted a phase 2 trial to assess the safety and efficacy of a single intravenous infusion of CIS43LS against P. falciparum infection in healthy adults in Mali over a 6-month malaria season [...]
At 6 months, the efficacy of 40 mg of CIS43LS per kilogram as compared with placebo was 88.2% (adjusted 95% confidence interval [CI], 79.3 to 93.3; P<0.001), and the efficacy of 10 mg of CIS43LS per kilogram as compared with placebo was 75.0% (adjusted 95% CI, 61.0 to 84.0; P<0.001).
This is a big deal: “In 2020 there were 241 million cases of malaria worldwide resulting in an estimated 627,000 deaths.”
SWURL Multi-Search 🔦
That’s a neat little one. I’ve bookmarked it on my phone.
Type one search query, and then swipe between resources:
Swurl.com is a web search engine that is optimally designed for mobile.
It instantly searches Google, LinkedIn, Instagram, Amazon, YouTube, Images, News and Reddit, and all results are viewed by scrolling or swiping — no clicking. Definite timesaver.
h/t Claudia Dawson
🎨 🎭 Liberty Studio 👩🎨 🎥
🎸 Senra’s Music Corner 🎶
My friend David Senra (📚🎙) has released a bunch of great podcasts on musical artists. I want to highlight a few of my favorites: