When everyone is betting the house on AI, there are probably some good deals on the other side of the line. Plus, the real problem with liquidation preferences.
The company that sticks out to me in today’s environment with large companies making large bets AND major changes around LLMs is Apple. Apple has the large ecosystem to indirectly benefit from people using their App Store and devices - but only barely - not nearly as much as AWS, Azure, etc. They didn’t pivot from their VR headset plan. They went ahead with what they had planned. We have given companies a hard time for years now about not being agile enough - but I think we are seeing some companies be too agile and not nearly focused or patient enough.
What other predictions you got, and which stocks should I buy?
(And this is probably naive, but things like this seem like they'll end up being a lot more useful than a lot of the foundational models. Right now, the assumption seems to be that models are, in a very rough sense, good as general application backends. But I could see models over time becoming a little bit more like employees, where they work much better when they're deeply integrated into the rest of the stuff they work with.)
That's a good question. I have no idea, to be honest. I would guess usually no, though it might show up in the fine print of some giant legal document.
I suspect you are right, and I suspect that the fine print merely discloses that liquidation preferences are possible, vs. the specifics that have been granted.
That's possible. It's also really messy (with or without liq prefs) to figure out, "If the company IPOs for $1b, how much do I make?" It takes a team of lawyers, a giant excel file, and a bunch of waterfall analyses to do it, and even that is often both a moving target and pretty error prone.
Sure, messy, of course. Not just liq prefs, either. All sorts of variables in valuing equity comps future value. And so I wouldn't expect anyone to predict the future, but I think it's reasonable to disclose that these conditions exist with actual investors, and that liquidation preferences have an impact on the future value of my options/shares.
I've been at enough startups to know that the hype machine when you're being courted as a prospective employee is set to 11, but the disclosure dial is around 2 or 3. And often if you ask about a specific detail, it's not known, or poo-poo'd as not important, when in fact it is.
The information asymmetry is high between startup and employee. When I was younger my attitude was 'let's work hard, go conquer the world, and make a lot of money.' Now my attitude is 'thanks for the pile of lottery tickets, let's work hard and maybe something will come of it.'
Yeah, I generally agree. There are often inevitably complications for a lot of these things, most companies try to sell the hype to employees as much or more than they do the market. Which, on one hand, I get - there's some element of, you'll only be able to build it if you believe it. But on the other hand, I would imagine that most of the FanDuel folks had no idea what was going on.
lol - nope! I was actually congratulating Apple for sticking with VR and NOT just jumping on the bandwagon - moving resources around because LLMs are hot. However at this point, how could they not invest in them - or atleast say they are…
Agree - I think they will be like employees on slack you can chat with in group chats or DMs that can do various things.
The thing I want to see more of is the model knowing when to ask for clarification / input / logical next steps!
(I do think this makes AI video potentially more interesting than text, because 1) training data can be not-really-human generated, and because 2) you can passively collect tons of it. Neither of those are quite true with text)
The company that sticks out to me in today’s environment with large companies making large bets AND major changes around LLMs is Apple. Apple has the large ecosystem to indirectly benefit from people using their App Store and devices - but only barely - not nearly as much as AWS, Azure, etc. They didn’t pivot from their VR headset plan. They went ahead with what they had planned. We have given companies a hard time for years now about not being agile enough - but I think we are seeing some companies be too agile and not nearly focused or patient enough.
Ha! Then Apple goes and launches MM1. So there is that.
What other predictions you got, and which stocks should I buy?
(And this is probably naive, but things like this seem like they'll end up being a lot more useful than a lot of the foundational models. Right now, the assumption seems to be that models are, in a very rough sense, good as general application backends. But I could see models over time becoming a little bit more like employees, where they work much better when they're deeply integrated into the rest of the stuff they work with.)
I wonder how often liquidation preferences are disclosed to employees?
That's a good question. I have no idea, to be honest. I would guess usually no, though it might show up in the fine print of some giant legal document.
I suspect you are right, and I suspect that the fine print merely discloses that liquidation preferences are possible, vs. the specifics that have been granted.
That's possible. It's also really messy (with or without liq prefs) to figure out, "If the company IPOs for $1b, how much do I make?" It takes a team of lawyers, a giant excel file, and a bunch of waterfall analyses to do it, and even that is often both a moving target and pretty error prone.
Sure, messy, of course. Not just liq prefs, either. All sorts of variables in valuing equity comps future value. And so I wouldn't expect anyone to predict the future, but I think it's reasonable to disclose that these conditions exist with actual investors, and that liquidation preferences have an impact on the future value of my options/shares.
I've been at enough startups to know that the hype machine when you're being courted as a prospective employee is set to 11, but the disclosure dial is around 2 or 3. And often if you ask about a specific detail, it's not known, or poo-poo'd as not important, when in fact it is.
The information asymmetry is high between startup and employee. When I was younger my attitude was 'let's work hard, go conquer the world, and make a lot of money.' Now my attitude is 'thanks for the pile of lottery tickets, let's work hard and maybe something will come of it.'
Yeah, I generally agree. There are often inevitably complications for a lot of these things, most companies try to sell the hype to employees as much or more than they do the market. Which, on one hand, I get - there's some element of, you'll only be able to build it if you believe it. But on the other hand, I would imagine that most of the FanDuel folks had no idea what was going on.
lol - nope! I was actually congratulating Apple for sticking with VR and NOT just jumping on the bandwagon - moving resources around because LLMs are hot. However at this point, how could they not invest in them - or atleast say they are…
Agree - I think they will be like employees on slack you can chat with in group chats or DMs that can do various things.
The thing I want to see more of is the model knowing when to ask for clarification / input / logical next steps!
They're trained on a bunch of Reddit posts, so I assume they're doomed to confidently say nonsense forever, unfortunately.
Training AI on human generated content - that was the first mistake...
(I do think this makes AI video potentially more interesting than text, because 1) training data can be not-really-human generated, and because 2) you can passively collect tons of it. Neither of those are quite true with text)
Hahaha. Ya… yikes.
Good edition.
I’m with you, shut up and drive is a straight banger