superfrank

It's awesome that it had the capability to do that, but it worries me a bit that it chose to do that.

I just listened to a podcast last week with the author (Dex Horthy) on as the guest. Humanlayer (his company) is constantly trying to push the limits with agents and he's very into using them.

He's well aware of exactly how good the models are today and I'm sure that if he thought they were good enough now to make the dark factory idea work, they'd be trying again.

Just to offer a different perspective, my last job removed QA and it was a nightmare. Feature delivery slowed way down and incidents and bugs went way up to the point where we started losing customers.

In theory, I agree with what you're saying, but I think a codebase needs to be in a certain state and a company's leadership needs to be in a certain mindset to make it work. Our codebase was far too large and interconnected for any developer to fully understand all of it because of years of building with unclear boundaries between teams. QA's value wasn't really the testing they did. The value was that we had a dedicated team who's job it was to see the big picture and keep up with the changes 15 different teams were making on the same product. Management wanted the benefits of removing QA without the pain of paying down all the organizational and tech debt that made them valuable.

Again, I agree with your sentiment, but also, I think that most large organizations are not willing to do the necessary work to get a place where QA can be removed.

Fashion trends tend to be roughly on a 20 year cycle. Uggs are back in fashion with GenZ

https://www.elle.com/uk/fashion/what-to-wear/a69924640/how-t...

https://www.vogue.co.uk/article/how-to-style-ugg-boots

I don't really believe the idea that there's some grand plan by the Chinese government to use their AI labs to undermine the US companies. Maybe now it's more of a reality, but I don't believe it was there from the start.

That said, if were going down the rabbit hole of saying that the Chinese labs are now part of a larger geo political strategy by the Chinese government, then I think Taiwan is part of the equation. If the frontier models require the best chips and those are mainly coming out of Taiwan then it's hard to imagine a world where the US allows China to make a move on Taiwan without a fight. If frontier level models can be run on chips being made elsewhere, then Taiwan becomes less important geopolitically. I don't think we'll ever get to a world where the US is just like, "Fine China. You do what you want.", but China has to assume that there will be a lot less resistance from the US is Taiwan isn't such a key component in the AI race.

Based on what I first heard it described as, yeah, I think so.

In most places I've worked SAs are generally just connecting existing pieces of a system together to meet a customers needs. They may write code, but it's often just glue to connect two bits of the system together or transform some data or something like that. They're not really contributing to the underlying product, they're just using the product and some custom code to meet a specific customer's needs.

An FDE is supposed to be closer to a regular software engineer on a product or platform team in that their goal is to solve a specific customers problem, but they're supposed to be more focused on the big picture and using their learnings to build a better product. They're still using existing systems to solve problems and writing plenty of glue code, but they're also supposed to have the leeway to contribute to the underlying product to make it better for all customers.

The simplest example I can think of would be something like a customer saying something like they need a way to convert a bunch of their data to a CSV and then send that to a certain email address every Friday. A more traditional SA mindset may be to write a Python script that runs on a cron that connects to that customers DB pulls the data, converts it to CSV, and then emails that specific email address. Even if the SA knows that's not the best way to do it, it's the tools they've been given to work work. A FDE should have the leeway and skills to go talk to a PM and their engineering team and just build an in product, self serve tool to do that (assuming everyone is aligned that that is good for the product).

Again though, what I've heard is that most FDEs at most companies are just SAs by another name.

I agree that the idea is cool, but from what I've heard from people in the role at most companies it's essentially a solutions architect role by another name.

Funny enough, the Pragmatic Engineer (author of the post linked) had a follow up from about a year after the post above and he reports the same thing.

https://blog.pragmaticengineer.com/the-pulse-forward-deploye...

It's also possible that there are more in-depth detection methods and that this was just a cheap and easy first step that hasn't been removed because it catches a lot of less sophisticated bad actors.

It's unlikely that this will stop a big AI lab from distilling their model if they're really determined, but A) it may be enough to stop a bunch of fly-by-night token resellers looking to make a quick buck and B) you never know when one person at one of those big labs will mess up and forget to install whatever workaround they have and out themselves.

I think of it like if you have a problem with birds in your yard so you go buy one of those plastic owls. The owl scares away most of the birds, but not all of them, so you go and buy some ultrasonic noise thing to scare them away (I'm just making something up). Just because you bought the new ultrasonic thing though, that doesn't mean you're going to take the owl down. You leave it up because now you've got two layers of defense instead of one.

OpenCode has a $10/mo sub that includes both of those

I also hadn't heard of it, but I feel like it's kind of a corollary of the whole, "what's measured is managed" idea or maybe the Streetlight effect (https://en.wikipedia.org/wiki/Streetlight_effect).

It's easy to measure a doorman's cost, but hard to measure their impact. Few, if any, guest are likely to mention the impact of a doorman on their stay except in the exceptional case. That means when budgets start to get tight (or an exec wants to drive the share price up), doormen become an easy target to cut because there's little hard data to justify their value.

One of the people I know had that happen and didn't realize it. Between the exercise cost and taxes I think it ended up costing them nearly a quarter million.

I know that seems insane, but even at a $500m valuation their potential payout is enough to retire on.

I was an early employee (#20-something) at a company that peaked at just over a billion dollar valuation during Covid and that now sits at somewhere between 1/2 and 1/3 of that depending on who you talk to.

I'm still really close with a lot of early employees and while I was lucky enough to have a liquidity event happen shortly after I left that allowed me to cash out for a decent, but not life changing, return, many of my friends were not.

One of the things I think a lot of people may not realize is how badly this zombiecorn state fucks employees with stock options. A lot of startups will give you a limited amount of time after you leave to exercise your stock options (90 or 180 days is common based on my experience). If you don't exercise your stock options and buy your stock within that time period the options expire and you get nothing. The problem is that if you buy the stock, you won't be able to sell until there's a liquidity event (usually a new funding round or IPO) and current investors don't want to take investment at a lower price unless they absolutely have to.

I know some other early employees who were laid off who had to make the choice between dropping $75k or $100k to buy stock that is worth 10x that on paper (even at the current valuation) and praying for a liquidity event that will probably never arrive or letting go of shares that just a few years before seemed like they would be a life changing amount of money. I know people who've done both and neither route leaves people feeling good about their decision.

I know common wisdom is that you should treat that stock like it's worth nothing until they day you sell, but when you've worked at somewhere for 5-10 years and seen the on-paper value of your stock rise to a life changing amount of money, I think it's hard not to assume that you'll be able to cash that out one day.

> they profit off of peoples hopes and dreams

Not trying to be overly cynical, but isn't that just marketing?

A lot of marketing just getting people to picture a better future version of their life and then making them think that your product will get them there. They're not actually buying the product for the product, they're buying it to try to get that imaginary future. I don't really see how Repilit ad telling people they can built the app of their dreams is very different than a gym ad telling people that they can get ripped or something like The Container Store showing someone with a messy house magically getting organized and cleaned.

I'm not saying that any of those examples are particularly good or moral, but I don't get how what Repilit is doing is any different than just standard marketing tactics we see every time we watch a block of ads.

I've got a couple of LLM wikis running for different purposes. I just pointed Claude at Karpathy's Github Gist and said "do this" and it set up and has maintained them ever since. So far no issue with that.

Can you explain why the version linked is better?

Gist link: https://gist.github.com/karpathy/442a6bf555914893e9891c11519...

"It doesn't look like anything to me"

I don't think you appreciate how much time, money, and energy a really bad tenant and an eviction can cost.

> The world can't deal with something that requires global concentrated efforts

Historically, that's not correct. The Montreal Protocol to reduce CFCs in response to the hole in the ozone layer is a perfect example of us doing this.

I realize the world has changed and maybe it's not possible in our current political climate, but we have worked together as a planet to solve these type of global problems before.

> In theory landlords would be looking to petition for eviction but that's usually not what happens.

For a small time landlord, a good tenant is worth their weight in gold. It's not worth chasing out a reliable tenant you have a good history with just to try to squeeze a few extra bucks out of the property.

> If the number of PTO days taken is the same between the groups, isn't it CLEARLY superior from an employee perspective to have a set amount

For me, no. I take more PTO than most companies with limited PTO offer. Most big tech companies offer 20-25 days and I'm taking 30 minimum. I take 6-8 weeks a year and the year I got married, I took 10. A normal year of PTO for me is 2 weeks during the summer for a vacation, a week at Thanksgiving, a week in April or May since my family has multiple birthdays in that window, 2-3 weeks in the Christmas/NYE period, and then random days here and there for the rest of the year.

For my team, for the people on the low end banked PTO is probably better. For the people on the high end they often take more than the allotment for CA employees, so no.

Everyone's experience will be different, but in my career I've always found that unlimited PTO often means no one cares how much PTO you take as long as you're getting your work done. I value that freedom. I know it's not for everyone and I'm not saying that every company should be unlimited PTO, but I just hate this narrative that it's a scam.

> But adding all that manager and corporate discretion sets one up for abuse when things go wrong at either the manager or corporate level

I have never worked at a place where I didn't need manager approval for my PTO. That includes places where I had a set PTO balance.

In my experience, having banked PTO days doesn't actually give you any real protection from abuse. A manager can still deny every single PTO request or load you up with so much work that taking any PTO will result in you falling behind. The only difference is that the company then needs to pay you out for those days, which isn't nothing, but it's also not a ton of protection from abuse.

> But what happens in practice is no one feels like they are entitled to the time they should be entitled to, and negotiations from the employee side always come from a place of weakness. It's a terrible system

> Undoubtedly someone will respond to this post with just how amazing their manager is and that they have never had a problem.

That me! Except I don't think it has anything to do with my manager or company.

I've worked 5 different jobs over the last 12 years with 8 or 9 different managers and literally never had an issue with taking the time I want while taking 6-8 weeks of PTO a year. I've hit the point where when I'm looking for a new job unlimited PTO is kind of table stakes.

I manage a few teams now with some people in the US where my company does unlimited PTO and others in Canada where our company cannot give unlimited PTO. Looking at my teams, the amount of PTO people take has almost no correlation to whether they have unlimited PTO or a set number of days. I have US employees who take a ton of PTO and Canadian employees who have burned through their entire balance and then some and I have employees in both places who take essentially none.

I get that if you're in that second group it's preferable to be in a place where you'll get paid out for the days you didn't take, but I'm pretty convinced that unlimited vs set days has almost no bearing on how many PTO days someone will actually take.

I have thought for a while now that skills were a bad abstraction. There's a lack of definition around what to use them for that I think contributed to why they rose to the top, but that's also why I think they aren't a good long term option.

The fact that I can have a skill that is just general guidance on front end design best practices that an agent can call upon whenever they feel, and another that is essentially a run book of steps that need to be followed exactly only when explicitly triggered, and a third that is basically just instructions on how to use a specific tool and all of those are acceptable just feels wrong to me. I get why it caught on and why the flexibility is attractive when the entire world is collectively learning a new tool, but skills have come to feel like the junk drawer in the kitchen where you just throw random shit when you don't want to think about a better place to put it.

I would love to see the world standardize on something like:

- Agents: Essentially personalities for a model to take on. This becomes the new place for skills like "front end expert" where you're not telling an agent to do a specific thing, just to think in a certain way about a task.

- Prompts: Repeatable instructions for specific tasks that an agent should follow when prompted. This could be something like a checklist style run book on how to resolve a certain error that an agent needs to follow exactly or it could be something like here's an idea I have for a new feature please poke holes in it.

- Tools: Tools (like CLIs, MCPs, or scripts) and instructions on how and when to use them. I'm purposefully not calling this skills because I think the term is overloaded, but that's kind of what this is.

> Polymarket doesn't seem to care too much

Polymarket is now two separate entities. Last year they were banned from the US and stopped taking on new US customers and made their site withdrawal only for anyone based in the US. You could bypass it with a VPN, but they did enough to at least look like they were complying.

They've now created a separate company that I believe is called Polymarket US which has it's own version of the site/app which they are using to re-enter the US market. This switched from invite only beta to open sign ups a few months ago.

Polymarket US is mainly just sports betting with a small number of crypto, commodities, and political election based markets, although they have said they plan to expand their market options. Their US based site almost certainly doesn't violate those rules. The version that is used outside the US is the one that has markets that violate that, but since they're not in the US, there's not much that can be done.

I'm not an expert in this, but I thought one of the biggest arguments for why a wealth tax is needed the whole "buy, borrow, die" thing where the ultra rich can use their assets as collateral to take out a never ending series of ultra low interest loans until they die and then have most of the tax burden of selling assets to pay off those loans wiped out because the tax code is much more favorable to selling assets to pay off the debt of someone's estate.

If (big if) I'm remembering that correctly, I don't get why we just go after the problem directly and do something like treat putting down collateral for these type of loans as a taxable event. I'm sure it's not as straight forward as it sounds, but I can't imagine it'd be more convoluted that needing to track the wealth of every high net worth individual.

Maybe I'm in the minority on this, but I actually don't care if Jeff Bezos' net worth went up by $5 billion because Amazon had a good day in the market. If the shares are just sitting in an account doing nothing other than proving ownership it's all kind of just numbers in a computer, IMO. A painting is probably a better example than stock, but if I have a painting on my wall that was worth $1 million dollars yesterday and today it's worth $10 million that change in valuation is essentially meaningless as long as the only thing the painting is doing is hanging on my wall.

What I do care about is when he's able to access the cash value of that $5 billion of Amazon stock without paying the taxes that would come along with selling the stock. If he wants to leave $5 billion in Amazon stock just sitting in his account doing nothing until the day he dies, that's totally fine, but the second he puts it up for collateral we should tax that. I think this has the added benefit of simplifying things by avoiding a lot of questions around fair valuation of assets. If I have a $10 million dollar one of a kind painting on my wall that I'm never planning on selling, it's kind of hard to put a valuation on that and it can be easily manipulated by finding the right appraiser. If I put a painting up as collateral for a $10 million loan it becomes a lot harder for the owner to argue that it's actually worthless or the IRS to argue that it's actually worth $1 billion.

> > Each 1% of wealth tax is equivalent to 20% of income tax.

> Mathematically sound.

Don't most wealth taxes that have been proposed have a certain level of wealth that you pay no taxes on? If so, doesn't that make this at least partially incorrect?

Maybe I'm missing something, but if I have $100 and have to pay a 1% wealth tax on it then sure that's roughly 20%. If I have $100, but I only have to pay a 1% wealth tax on everything over $90 that's more like a 2% income tax.

The rules for each contract are provided when you bet, but ultimately there are plenty of markets that are settled in controversial ways and users have little recourse because the sites TOS's often say their rulings are final (this is the same for many sports books as well).

To give an example, I wagered on a market a while ago that Trump would say "Mamdani" before the end of the week. He responded to a question Mamdani where the reporter asked about the mayor by name and Trump said "Mandami" instead of "Mamdani" (switched the m and n). Kalshi ruled that that didn't count as Trump having said the word.

Trump ultimately said Mamdani correctly the next day so it ended up not mattering and I think the rules have since been updated to accept obvious mispronunciations, but I think it's a good example of how much gray area some of these markets can have.

I'm not saying this as an argument for or against prediction markets, but that's essentially what the vig is at traditional sportsbooks.

Someone calculates what they think the odds of an outcome happening are and then they allow people to take positions on either side at worse odds than what they think the real odds are. As long as their prediction is correct, over time they make money. It's why putting $1 on a 50/50 bet on a sportsbook will usually only pay out around $1.91 instead of $2 if you win.

90% of the betting volume on Kalshi is betting on sports. I know how prediction markets work and how they're different that traditional sportsbooks, but they ultimately allow customers to do the same things.

If you are saying Kalshi isn't a sportsbook because the house isn't on the other side of the bet, you might as well argue that DraftKings isn't a sportsbook because they don't actually track your wagers in a book.

90% of the volume on prediction market sites is on sports. Yes, there are smaller markets that are easy to manipulate, but that's no different from traditional sportsbooks.

Yes, What MrBeast says on his next video is easy to fix, but so is something like betting on a D3 basketball player to have less than 10 points or some English 5th league soccer player to have a yellow card.

Yep. I totally agree.

For the record, I'm actually a very active Kalshi user, but I cannot see how anyone can seriously argue that it's not sports betting.