Sunday Reads #166: If only Maverick had such a wingman...
Sunday Reads #166: If only Maverick had such a wingman...
Plus notes on shitstorms and laugh riots from the last week.
Hey there!
It’s been a long week on the Internet. Felt like a decade, hasn’t it?
Let’s talk about FTX and Twitter, sure. But first, let’s talk about something a little more… optimistic.
1. The best copilot you could ever have.
I read this astounding fact the other day, about Github Copilot.
If you turn Copilot on, then you need to write only 60% of the code yourself!
So how does Github Copilot work?
Copilot uses a technique called "generative programming". Which is a form of AI that helps you generate new code.
For example, if you're working on a piece of code, and you're not sure how to do something, you can ask Copilot for help.
It will then generate a new piece of code that does what you want, and you can use it in your project.
Now, Github Copilot is legit amazing! 2x-ing productivity straight off the bat.
But I realized - this is also a great framing for where we are today, in the evolution of AI.
AI as a copilot to make us better at everything we do.
And it all starts with the current revolution in Generative AI.
First, what is Generative AI?
Generative AI, as the name suggests, is AI that can help us generate new content.
You give it a prompt - could be text, images, code, anything - and it will complete the rest for you.
Some examples of generative AI include:
Generating new text (GPT-3)
Generating new images or videos (DALL-E or Stable Diffusion)
Generating new sounds or music (Resemble.ai, podcast.ai)
It's not just about automation. It's about becoming more creative. It's about getting more ideas.
How do these tools do this?
They use Large Language Models, or LLMs.
These are neural networks trained on humongous amounts of data (think - the entire Internet) to learn the statistical relationships between words. Or between words and images.
Once trained, they can generate new content given a prompt.
That's all. This simple (but also hugely complex) pattern prediction model can be used in a huge variety of ways.
In each of these areas, Generative AI tools are like a Copilot, gently making you better at what you do.
Copilot for writing (GPT-3).
I write a lot, so I've been able to give GPT-3 a spin. And I've got to say: it blows my mind how good it is, at knowing exactly what I'm going to say next!
In fact, a lot of this blog post was written by an AI tool (I used a GPT-3 plugin inside Roam Research).
Apart from using these to generate text, you can also use them to answer any questions you have. Because these models are trained on the entire Internet, it's almost like "asking your all-knowing friend a question".
For example, in the previous section, I just wrote down the term "Large Language Models", and asked GPT-3 to explain them in one line.
Sometimes, it gives better answers than Google.
The possibilities are endless. Writing marketing copy, cold emails, even creating your brand name and logo!
[Aside: it can even write jokes like Seinfeld. Although the voice delivery needs some work...]
But this just scratches the surface.
Because, a lot of other things can be reduced to text!
Writing code.
![A screenshot of a GPT-3 prompt and completion
Prompt:
Hi, GPT-3! Could you convert the following paragraph (in quotes):
"My name is Gabriella Gonzalez. I work at Arista Networks. My Twitter account is @GabriellaG439. One of the things that I'm working on right now is creating a bidirectional type-checking package and I'm kind of stuck on figuring out what the API should be."
… to a highly structured JSON expression?
Completion:
{
"name": "Gabriella Gonzalez",
"employer": "Arista Networks",
"Twitter": "@GabriellaG439",
"projects": [
{
"name": " creating a bidirectional type-checking package",
"status": "stuck",
"goal": "figure out what the API should be"
}
]
}](https://substackcdn.com/image/fetch/w_600,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fpbs.substack.com%2Fmedia%2FFanPkq3UIAEg1CE.jpg)
And this demo by OpenAI has several WOW moments.
They even created a game in 5 minutes flat, using just text!
Writing formulae on excel.
Check out the video in this tweet. A mind-blowing proof of concept.
Generating speech!
And this is just the tip of the iceberg.
Because what's next is even more powerful...
Copilot for images (DALL-E, Stable Diffusion).
The best illustration of the power of AI tools for image generation, is the video in this tweet:
There are now a lot of tools you can use - DALL-E, Stable Diffusion and Midjourney.
This is a great resource to get started: Getting Started With Stable Diffusion: A Guide For Creators.
And if you want to tinker right away, check out PlaygroundAI.
Aside: Because these tools can take both text and images as prompts, you can also use them to make yourself more photogenic!
For example, this is what I did:
Now, if only one of these photogenic guys could attend Zoom calls in my place…
[PS. If you want to try this too, go to ProfilePicture.ai. This is an affiliate link.]
Copilot for video.
At its most basic, video editing can be automated or abstracted. Instead of spending hours replacing video backgrounds / removing artefacts one by one, you can just... you guessed it... tell your computer what to do in plain English.
I haven't tried this yet, but you can also use Stable Diffusion to generate videos from scratch, with some prompting.
And there's this new tool Phenaki that I'm excited to try!
Copilot for everything.
Ultimately, what GPT-3, DALL-E, Stable Diffusion and others show us is this:
A lot of what we call creativity is just pattern matching.
And even if the entity doing the pattern matching doesn't actually understand the patterns... they look quite all right!
You might say, "spitting out patterns is not intelligence".
Well, I dunno about you, but this is what I do all day.
Case in point:
This thread lists 10 websites that are all use cases of Generative AI.
I'll touch on one of them, as an example: 9/ Namelix.
Namelix is a website that generates names and logos for your brand / product, in the blink of an eye.
I asked for a name and logo for a retail cashback app (which was incidentally the app I had created in 2014). And this is what I got - an *infinite scrolling* list of brand names and logos!
Not bad for 5 seconds work.
Looking beyond the corner.
Over time, AI will become more powerful. It will take over more of what we do. And it's even odds whether that'll make us more effective, or more redundant.
But until then, think of it like that potion Super Mario drinks, that makes him 10% better at what he does.
Looks like we'll have a fun couple of decades. Before our superhuman AI overlords gain sentience and kill us all, anyway.
Before we continue, a quick note:
Did a friend forward you this email?
Hi, I’m Jitha. Every Sunday I share ONE key learning from my work in business development and with startups; and ONE (or more) golden nuggets. Subscribe (if you haven’t) and join nearly 1,400 others who read my newsletter every week (its free!) 👇
2. The shitstorm that is FTX.
From a valuation of $32B on Monday to bankrupt on Friday. Over USD 8Bn of consumer funds lost. To hubris, stupidity, and some theft too.
Hope you were not too affected by this. An expensive reminder of the old crypto adage: "Not your keys, not your coins."
Trung Phan has the run down, in FTX: The $32B implosion.
As you read all the obituaries of FTX, notice:
The signs of something hugely amiss seem to be in plain sight.
SBF and his team always on drugs
Incompetent and inexperienced risk managers
Convoluted org structure
SBF not even good at League of Legends!
But also notice:
No one. NO ONE noticed these obvious signs before.
I'm sure they existed. But not a single critic voiced them loud enough.
Instead, they were praising SBF to high heavens for his brilliance and selflessness.
Sequoia wrote a 13,800 word (!) hagiography of the guy. It would be worth reading. If only they hadn't taken it down yesterday.
Luckily, I saved you some of the sauciest quotes in the thread below 👇.
This is what herd mentality and extreme hero worship look like. Remember this.
3. The laugh riot that is Twitter.
Not all is doom and gloom on the Interwebs.
Elon's takeover of Twitter is progressing exactly as you'd expect. Pure chaos.
The $8 verification which was rolled out in "move fast and break things" fashion, well... broke things. A lot of things.
On the plus side, we got to see some fantastic tweets, from "verified" accounts.
At least in one case, someone paid $8 for Twitter Blue and wiped out several billions in market cap.
Oh wait, there was one more.
Too bad they've now rolled back these verifications.
I'm rooting for Twitter to survive, and I'm rooting for Elon to pull it through.
But in case he can't, at least he's providing us a lot of laughs along the way.
That’s it for this week. Hope you enjoyed it.
It’s been a rough week, and layoff season still isn’t over. As always, stay careful out there.
I’ll see you next week.
Jitha