> "Lean-checked" is just another quality improvement, like reviewing your code carefully or adding some unit tests. It's not—it's totally different
I think for most things, this is actually a good way to see it: As a third leg on top of reviewing your code carefully and unit tests.
As you say, people will still make many mistakes in how they write their formal requirements, or miss things they care about.
For math this is not an issue, it's usually easy to check that the lean statement is exactly what you want. But for software, and many other real world applications, it's better thought of as a new orthogonal testing vector that everyone should be using.
> But for software, and many other real world applications, it's better thought of as a new orthogonal testing vector that everyone should be using.
Not sure about should, but as long as even modestly complex systems like the Signal communication protocol map into millions of lines of Lean, its pretty clear that not everyone will be able to use it. We still need better tools (tools better suited to our monkey brains).
Yes, I pulled the trigger to get a Google AI sub so I've been using it today. It's pretty decent, but dear lord it is slow. It's bad enough that I'm already thinking about cancelling.
Google’s segmentation here is keeping them from using their existing account dominance to eat away at OpenAI and Anthropic personal subscriptions, tbh. Adopt a lower usage limit for Carbon on the $20 plan and it’s suddenly insane value.
Given you get YouTube premium, file/photo storage, and $10 a month back in GCP credit which you can use on more tokens (including carbon). It already feels like great value to me, that said how much you get out of it will depend how deep you're in Google's ecosystem.
YouTube Premium is only personal so it's useless for me I'm already paying for family plan, and what do I need 20TB of cloud storage for to update to 100$ plan - basically it's a useless bundle for me not a value proposition beyond the 20$ with their usage limits I saw.
Same for me. I find myself using Google’s $20/month plan, plus running really good local models with occasional use of Deepseek Harness with their flash model, occasionally a $6/month Alibaba subscription for their strong qwen-3.8-max model, and about every third month getting a Anthropic subscription if I need more access to Opus.
This is a lot of purchasing complexity for an old retired guy like myself!
I wish Google had finer grained $30/month and $40/month subs: then I would probably just make my life easier by only using local models and Google (their sub also provides some Opus, but the Anthropic subscription is better if all you care about is Opus or Fable.)
I find Lite to be perfect. It only adds ads to copyrighted music videos and doesn’t include YouTube Music, otherwise it’s the same as YouTube Premium. For $100/yr with lots of cloud storage and Gemini access, it’s one of the best value subscriptions out there in my book.
AFAIK the real issue, and also driving the pricing structure for API and the "50% discount on new model" thing, is frankly capacity of the TPUs they have. Not enough TPUs, not helped by renting some of them to Claude (lots of enterprise users there, so at least it pays to GCP, eh?) and now OpenAI
Very cool to see this kind of hypothetical working in practice.
I wonder if it's also possible for agents to communicate via a shared KV-cache. Maybe by spawning sub-agents with carefully crafted system-prompts/prefixes. KV-caches are often shared between many computers, so it would be a powerful communication channel.
> Noteworthy: ironically, OpenAI and Anthropic's models refused to do this work. Open weight models got the job done. I'm worried the labs' stance might backfire and get in the way of helping developers build correct and secure systems.
I was working on a network protocol and it correctly found a DoS attack. Once found - Claude refused anything and everything to do with it. Even asking what it was reading locked immediately. I had not written any of it.
Sol 6.1 only denied me once.
Having a multi-model harness was always the better choice, but this really drove home the point to not get too dependent on any single provider.
I've been working on a game server in rust, and every once in a while a review will just stop and refuse to explain what happened. So I know there's a security issue in there somewhere, but neither OAI nor Ant models will tell me what it is. Despite the whole thing, top to bottom, having been built in the last two months on the system they're working on.
Planning to hop on openrouter this weekend to chase it down, but man, it's rough knowing I've already burned the tokens to find it probably half a dozen times at this point, but don't have any information that would help point me at it.
That seems like the least charitable reading of what I wrote.
"system they're working on" refers to the actual physical hardware.
Yes, I look at the code regularly, and a significant portion is hand written, especially the performance critical sections.
A security issue I haven't been able to track came up in automated code reviews. Maybe you're such a savant that you can crack any security issue in a codebase you're working on, but the last few months of critical security bugs in infrastructure with the most experienced and capable maintainers in the world should at least humble you a little.
The other day in one of the Android TV related threads somebody posted a Github repo for an AndroidTV browser app with a rich suite of functionality. I pulled the repo down and asked Claude to scan it because I was concerned about security issues and vibe-coded vulnerabilities. It got about half way through the scan and abruptly stopped, refusing to scan it more nor would it say exactly why it stopped or what it found, only that it was a "security" issue.
Needless to say, I ditched the repo. The problem is I don't know if what it found was legitimate or simply a keyword that set off alarm bells or if my prompt was falsely detected of being malicious. It's very frustrating. When I researched more on this topic, I found users complaining about the same issue when asking Claude to scan new releases of legitimate npm packages (a popular library like react-query for example).
They've simultaneously cultivated a medium for faster and more devastating exploits with no good way of protecting ourselves from them. It's ridiculous.
The real question is: what's the rate of growth for capital returns vs labor returns? If capital returns are accelerating, over 50-100 years, yeah, we will still end up in techno feudalism.
If you scale a model down, or switch out attention with linear attention, or introduce KV-compression, etc., you don't really know how it will affect model performance before you've trained the model.
And once you've spent millions training it, you might as well make it available if it's pareto optimal for some tasks. It also gives you more insight seeing how people use it in practice.
"Unless you have special insight, you should assume all stocks are equally good buys."
Lol maddest thing I've read all day
"priced in"
I smell your efficient market hypothesis argument, before you even begin just keep in mind that does not apply here, it's an IPO, the price is not set by the secondary market
> "Unless you have special insight, you should assume all stocks are equally good buys."
> Lol maddest thing I've read all day
didn't someone like the WSJ to a long term experiment picking a basket of stocks by throwing darts at the paper? They beat most/all established funds over 5 years or something like that. I don't think it matter much, i just buy VOO when i get my bonus and go back to living my life. That strategy is on track to put two kids through college.
Regarding the stock picking experiment, yes, there have been many similar experiments. The premise is just pick a random basket of stocks and see how it performs over decades. Indeed those have outperformed cap-weighted the indexes, most of the time.
How? If you pick a random basket of stocks, you'll end up with mostly small-cap stocks. Cap-weighted indexes are heavily driven by a few large stocks. Most years, small-cap outperforms large-cap (Fama-French 5-factor model[1]). So the "monkey picks stocks" experiment is actually just an exercise in harvesting what is called the "size premium".
Since the AI boom, large-cap has been significantly outperforming small-cap, so if you run the same "monkey picks stocks" experiment, it would likely underperform most of the time.
If you want to easily visualize this, just go on TradingView and compare SPY (cap-weighted S&500) with RSP (equal-weighted S&P500).
I would actually argue that if they price the IPO too low, people would assume they are not confident in their company and the whole AI sector would sink.
Almost like a "Veblen good" type investment.
They basically have to price it at an unreasonable valuation so that people curse them while they take it. There's no way around it.
I think the argument is that the valuation is so high because the enormous growth IS priced in, but still being far from guaranteed. Meaning there's only downside: if the enormous growth is realized it would be a wash for your investment since it is already priced in, and in the plausible scenario it fails spectacularly or only grows moderately your investment will lose a lot of value.
It's an IPO, nothing is "priced in" until is actually being transacted in the secondary market
There's a good reason IPOs are shown to underperform the market
Yes, lots of downsides on this one, but people see it as a call option on a future where AI replaces all jobs, what would you pay for that? A rational response from someone that accepts the premise of the question is likely "anything"
> It's an IPO, nothing is "priced in" until is actually being transacted in the secondary market
Well, priced into the IPO price by Antrophic. It was you who said it would be priced into the IPO at launch, I just used the same terminology you did: "There's not going to be an IPO at a price where everyone agree that it'll see a 30% growth over 30 years. Obviously that'd be priced in."
If you don't think one should assume the market's expectations of future results are accurately priced into the IPO price, it's an argument in favour of the comment you tried to argue against, not contrary to it.
> Yes, lots of downsides on this one, but people see it as a call option on a future where AI replaces all jobs, what would you pay for that? A rational response from someone that accepts the premise of the question is likely "anything"
Even if you buy that premise, there are more premises needed. First (1) AI replaces all jobs, then (2) private AI models will be able to convert it into profit without being contested by open models, razor thin margins due to competition, regulation/nationalisation, or general societal upheaval, finally (3) Antrophic specifically will "win the race" and capture the lion's share of that market.
That being said, I would not bet any money on the stock losing value in any reasonable time frame either. It's enough to take a quick look at Tesla to see that fundamentals or actual results does not necessarily matter for a hyped sci-fi stock, even in the long term.
If that's what you think "priced in" means, I've a brand new company with a 1 trillion $ valuation priced in, would you like to buy some shares from me?
> It was you who said it would be priced into the IPO at launch
I actually said in my initial comment that the "priced in" argument is not even worth invoking, but here we are.
> If you don't think one should assume the market's expectations of future results are accurately priced into the IPO price, it's an argument in favour of the comment you tried to argue against, not contrary to it.
The comment I answered to initially said "Unless you have special insight, you should assume all stocks are equally good buys.", written by someone else. To which I replied "I smell your efficient market hypothesis argument, before you even begin just keep in mind that does not apply here, it's an IPO, the price is not set by the secondary market"
> (2) private AI models will be able to convert it into profit without being contested from open models, razor thin margins due to competition, regulation/nationalisation, or general societal upheaval, finally (3) Antrophic specifically will "win the race" and capture the lion's share of that market.
Sorry, I confused you with the user thomasahle that I initially responded to, which explains the confusion here. I don't think we disagree with each other.
It is my belief that while humans exist, there will be markets, one way or another. There will always be exchanges of value between people. There will always be people at different levels, and a form o currency to qualify for the higher levels. Even in communist societies this was true
So, a future where there is no market can only be a future where humans are extinct, in my opinion (some form of communism won't work)
If someone believes markets will still exist (one form or another) and there will still be exchanges of value between people, even in the most maximalist AI future - then I think it becomes rational for said person to seriously consider buying AI labs, even at stupid valuations
Admittedly I've formulated my previous comment wrong, I should've written "a future where AI has the potential to fulfill all jobs known today", instead of "a future where AI replaces all jobs"
As mentioned in other comments, I'm not trying to convince anyone to do anything, I'm here just to argue different points and positions
Production of surplus value is what generally qualifies as a "job" — I certainly agree there is potential for tools to replace/augment any job we think of, but we'll simultaneously continue inventing new jobs.
Just like economies are mostly a service-industry today, a concept that barely made any sense 100 years ago.
Still, buying tooling companies' shares makes sense when that company has something which is hard for anyone else to replicate: but we are seeing that it really is not with a number of proprietary and open weights competitors who generally go head-to-head with minor differences in capability.
This only to say that believing valuations are unreasonable does not mean you can't even be an "AI maximallist" — but markets do not work in a simple cost/value-provided equation for a long time now.
But in the past, rules & regulations basically forced companies to IPO at a certain size. This meant that when they IPO'd, there was still growth upside.
Those rules are gone now, so they can stay private and have all the growth happen privately. When they IPO now, it's more just an exit. The growth was captured by insiders only.
If that's your interpretation of things, that there is no logic behind some of the smartest people in the world throwing tens of billions at Anthropic with no logic behind it, I have no further comments
A general statement I made is mostly to refute that people invest exclusively when the data is clearly supporting that investment with guaranteed returns: basically, the fact that people will buy does not mean that they have guaranteed returns.
Some might be looking to capitalize early and exit quickly; some are creating demand because they have other drivers in their portfolio (eg. Nvidia or Micron); some might really believe in the company; and some might be gambling with their disposable investment money for their high-risk/high-gain part of portfolio.
Well, you're not refuting anything because no one claimed "that people invest exclusively when the data is clearly supporting that investment with guaranteed returns."
What you're attempting to do is called the straw man fallacy and it's a cheap shot.
they were not forced but they were forced to report like a public company even if they didn't IPO. all the cost with none of the benefit was an incentive to IPO earlier.
yes tho. once you hit a certain number of shareholders (which anthropic clearly is past), you used to have to report with the regularity and rigor of a public company. so if you do, why not IPO and get some cash?
that and other changes that enabled private investment at this scale. it's all about privatize the pump socialize the dump via IPO
The rule is still in place, if a company has more than 2000 shareholders, of which minimum 500 are non-accredited (employee stock based compensation does NOT count), the company has to do ongoing financial reporting
But those other ones at least have the room to potentially do that, even if most won't. So what GP is saying is that this is a magical stock indeed, but the other way around, in that it's magically capped.
reply