seanlinehan

Agreed, and I think this is reasonable. There are a great many examples of people who amassed wealth in unsavory or illegal ways. And my sense is that almost nobody defends all forms of wealth acquisition as moral and good.

That said, I think there are a great many people who believe that all forms of wealth acquisition (above a certain scale) are fundamentally immoral. And that belief is held because they believe, "The only way you get a billion dollars is taking advantage of working people, not earning it."

So, my read is that PG's point is not to defend all billionaires. But instead, to debunk the incorrect folk-belief that all billionaires necessarily are exploitative.

> if the economy is growing at 2.5%, how do you sustain 15% over 5 years?

US GDP is $31.82 trillion dollars per year. Taking the 2.5% growth rate, that's nearly $800 billion dollars per year in new GDP.

The economy very obviously does not progress as a bunch of soldiers marching in a straight line. Some firms will shrink 100%, some will growth 10,000%. This much is obvious by just looking around. But even if no businesses shrank, no wages were docked, nothing bad happened... even still there would be $800B in more GDP.

So if the economy is growing at $800B per year, it's extremely obvious how a company could even grow from $1M to $1B in revenue per year without doing anything shady... Just capture some of the new economic activity that cropped up this year!

And it's even easier when we're talking about an entrepreneur's net worth. Their net worth is going to be mostly holdings in company stock. The value of company stock is some multiple of the company's theoretical future financial earnings.

So if a company is making $1M revenue today, and growths to $5M revenue by the end of the year (15% MoM growth), at let's say a 30% EBITDA margin, they have made $1.5M EBITDA. And let's say that fast growth is rewarded at an extremely rich 50x EBITDA multiple. That company is now worth $75M. If this founder is lucky and owns 50% of their business, they now are "worth" $37.5M.

If they were only at $1M * .30 * 50 * 0.50 = $7.5M net worth at the beginning of the year, and then were at $37.5M at the end, their net worth increased by 500% in one year! And all they had to do was capture $4M / $8000M = 0.05% of the increase in GDP.

Like, none of this is either shady or complicated.

I think that's fair for certain one-shot generations. For example, sending off a single prompt to an image or music generator and just accepting the output.

But I think most of this stuff is iterative, multi-turn. You type a thing in, see what comes back, and then repeat until you have something that satisfies your desire.

Taking the manager analogy. If you spent 30-minutes writing up a draft spec, waited for the outputs, had a review meeting where you provided good feedback, and then repeated that cycle until the product was done... Again, I think that manager (assuming, of course, that their feedback was useful) should feel some pride in shaping that output!

One can reason by analogy here.

In a pre-LLM world, a classic software team would have PMs, designers, and engineers.

Of those three, the PM wouldn't have any real role in writing code. And they would rarely contribute a ton to the design. What they would be contributing is ideas, market insights, coordination, prioritization, etc.

When the product ships, one would expect the PM to feel a real sense of accomplishment. They helped this idea become a _real thing_! All of that pride, despite not writing a single line of code nor polishing any pixels themselves. And I don't think anybody would reasonably look down on them for that feeling.

Same thing with using LLMs. Sure, you didn't write the code. But you caused the thing to exist! That's exciting!

I would assume that at least one of two things are true:

1. They're costs are so so out of control that they need to impose a blanket cap immediately. Figuring out an allocation mechanism that can be deployed company wide is time consuming and they need to staunch the bleeding immediately, despite it being obviously suboptimal.

2. The few people who should have unlimited tokens were given exactly that. No reason to introduce such nuance to a public PR move. The hard-cap limit is a great negotiating posture with token providers.

Small models are super useful. But I'm skeptical of their use for coding in particular, which is what this model is advertised for.

Also didn't say that. You're arguing phantom arguments I very clearly didn't make.

The top 0.01% still pay enormous taxes. Elon one year personally paid $11B in taxes.

I get that a lot of people think people's unrealized capital gains should be taxed, so maybe the argument you're making is something like:

"People with very large paper-gains based on appreciation of the market-value of the assets they own pay 0% taxes on those unrealized gains"

In which case, yeah, that's definitely true. But if they sell those assets, they pay taxes. Some of the taxes from those sales can be offset by doing things like donating enormous sums of money to charity. And sometimes people take loans against their equity, which is not a taxable event. Though, in order to pay those loans back, they have to sell something (taxable) or earn money elsewhere (also taxable). So loans are tax deferral...

But eventually the tax man comes for everybody.

This perception that "lower ranks" are becoming poorer is just empirically not true.

On every metric, people in all income brackets are earning more on both a gross and COL-adjusted basis. It is the case that top quintile income has increased more than bottom quintile income, but a faster relative increase does not mean the other group is getting poorer.

The other very interesting thing is that there is statistically not really a "upper ranks" and "lower ranks". The majority of people in the 1% each year are there for the first (and often only) time. And a very, very small percentage of people in the bottom percentiles remain there for their whole life.

Some interesting research:

* 12% of the population will find themselves in the top 1% for at least one year

* Nearly 70% will spend at least one year in the top 20%

* More than half will have at least one year in the top 10%

* While 12% may reach the top 1% at some point, a mere 0.6% stay there for 10 consecutive years

All of that is to say, the idea that there are is some entrenched upper class waging war against some entrenched lower class is just empirically not true. If you dig through the data what you'll find is:

1. People who are just entering the workforce don't make a lot of money

2. As people spend time in the workforce, they make increasingly more money

3. When they retired, they start making less money but tend to have assets to live on

It's far more dynamic than most people's intuition leads them to believe.

I didn't say any of that. Taxes are fine.

The top 1% of income earners pay 40% of all the federal taxes collected. The top 25% pay 89% of taxes.

Net of transfers, 60% of households receive more from government transfers than they pay in taxes.

The idea that rich people don't pay taxes is just not correct. The entire system is basically rich people subsidizing everybody else through byzantine distributional systems.

This is definitely the way. There are good use cases for real sandboxes (if your agent is executing arbitrary code, you better it do so in an air-gapped environment).

But the idea of spinning up a whole VM to use unix IO primitives is way overkill. Makes way more sense to let the agent spit our unix-like tool calls and then use whatever your prod stack uses to do IO.

Reflexivity is nodded to in the definition of complex systems in the piece!

I think what you're saying is poverty is actually simple, and the solution is to stop the bad actors causing poverty? But at the same time, you are correctly recognizing that attempts to stop bad actors from causing poverty triggers reflexive responses and cascading repercussions. Which sounds mighty like a complex system?

Super interesting, I've never heard of this before. Thanks for sharing!

True -- I didn't mean to communicate that Santa Fe was a failure writ large. Their contribution was very important!

Though I think it's fair to say that the torch was picked up and carried by others with a different set of strategies.

Not the intention at all. The part about mechanistic interpretability was meant to gesture at how building such systems can provide new tool kit for building further intuition and understanding.

Agreed!

Perhaps a failure of communication -- I was indeed attempting to say that Chomsky was wrong and his ideas were interesting, but more or less a dead end.

I have this little bookmarklet in my bookmarks bar that I use constantly. It removes all fixed or sticky elements on the page and re-enabled y-overflow if it was disabled:

javascript: (function () {document.querySelectorAll("body *").forEach(function(node){["fixed","sticky"].includes(getComputedStyle(node).position)&&node.parentNode.removeChild(node)});var htmlNode=document.querySelector("html");htmlNode.style.overflow="visible",htmlNode.style["overflow-x"]="visible",htmlNode.style["overflow-y"]="visible";var bodyNode=document.querySelector("body");bodyNode.style.overflow="visible",bodyNode.style["overflow-x"]="visible",bodyNode.style["overflow-y"]="visible";var nodes=document.querySelectorAll('.tp-modal-open');for(i in nodes) {nodes[i].classList.remove('tp-modal-open');}}())

Some companies do things like this, but I'd be cautious about it.

There's a few things I'd consider:

* If you have a bunch of reps, doing the periodic accounting to cut the right checks becomes more of a pain (though it's a pain anyways)

* When you give a choice, the employee might make what, in retrospect, winds up being the wrong choice. This can lead to pissed off sales people and regrettable churn.

OTOH, sales comp plans change every year anyways, so could just be renegotiated

A simple salary + percent commission is a great model.

That said, this calculator was built to model/simulate the things that are super common in enterprise SaaS:

1. It takes sellers time to ramp up. Experienced sellers might be willing to jump to your company, but not if they are guaranteed to only get their (relatively) low base salary for 1-2 quarters.

2. If you decide to do a ramp, you have to make a choice about the OTE.

If you can avoid doing these things, that's great. Though whether that will fly largely depends on whether your sales cycle and target talent market supports it!

All of this is true, yet somehow misses the entire point of the post.

One of the best ways to get good at delivering actionable feedback is to practice a bunch of times!

It's slow and tough to get enough reps in real-life. Eventually, you'll get there.

But you can accelerate your learning by doing AI roleplays with a product like https://www.exec.com/ai-roleplays

In the early days, we bailed out to manual resolution. In the later days, we had enough disparate data sources that we built oracles to choose which of the conflicting data was most likely to be correct.

For example, we integrated with a data source that used OCR to scan container numbers as they passed through various way points while they were on trains. The tech wasn't perfect. We frequently got reports from the rail data source that a train was, for example, passing through the middle of the country when we knew with 100% certainty that it was currently in the middle of the pacific ocean on a boat. That spurious data could be safely thrown out on logical grounds. Other cases were not as straightforward!

Spot on.

I was an early eng and first VP of Product at Flexport. Global logistics is inherently complicated and involves coordinating many disparate parties. To complete any step in the workflow, you're generally taking in input data from a bunch of different companies, each of which have varying formats and quality of data. A very challenging context if your goal is process automation.

The only way to make progress was exactly the way you described. At each step of the workflow, you need to design at least 2 potential resolution pathways:

1. Automated

2. Manual

For the manual case, you have to actually build the interfaces for an operator to do the manual work and encode the results of their work as either:

1. Input into the automated step

2. Or, in the same format as the output of the automated case

In either case, this is precisely aligned with your "reuinifying divergent paths" framing.

In the automated case, you actually may wind up with N different automation pathways for each workflow step. For an example at Flexport: if we needed to ingest some information from an ocean carrier, we often had to build custom processors for each of the big carriers. And if the volume with a given trading partner didn't justify that investment, then it went to the manual case.

From the software engineering framing, it's not that different from building a micro-services architecture. You encapsulate complexity and expose standard inputs and outputs. This avoids creating an incomprehensible mess and also allows the work to be subdivided for individual teams to solve.

All that said – doing this in practice at a scaling organization is tough. The micro-services framing is hard to explain to people who haven't internalized the message.

But yeah, 100% automation is a wild-goose chase. Maybe you eventually get it, maybe not. But you have to start with the assumption that you won't or you never will.

It was not obvious at first glance to me, but the actual app is here: https://hn.wilsonl.in/

Site is nice! And it's super fast which is awesome!

Why not link to Amazon with an affiliate link in there? Collect a few bucks :-)

Link to the ImageBind site: https://imagebind.metademolab.com/

I ran an ISA company for a while and can say with pretty high confidence that there are many reasons why fee-for-success is not an effective way to monetize this type of content (Which I've written about before [0]).

I agree that a business, like YC, that has been around for a while, monetizes directly with a fee-for-success model, and has good reviews is very likely to be effective. But the inverse does not hold: a program using fee-for-service is not particularly good evidence that the program is ineffective.

[0] https://twitter.com/seanlinehan/status/1505923007096700939