lz400

we could even just repurpose the same office, "weights and measures" is oddly relevant

Yeah I don't get it. Why the maximalist hyperbole? AI is going to kill ALL humans! That's crazy. Why can't they warn about increased likelihood of cyberattacks or scams being easier to perpetrate like normal adults talking in public?

I don't know how to read this and not see that this is a direct accusation to OpenAI of having used the researchers data to try to front run his discovery on purpose. The evidence is not completely proven and also circumstantial but to me at least looks like a fairly suspicious situation.

I've said this before in here but Zitron is completely captured by his audience at best and a grifter at worst. A couple of months ago he was tweeting that people were crazy talking about _agents_, that they didn't _exist_ and that people talking about them were shills or bots. The responses were full of incredulous software developers saying but but but I use them every day, they're so good they're scary actually...

Zitron is just, like Dan Luu says, wrong about everything and doesn't care anymore, he's in the business of extracting money from his engagement.

It’s sketchy to invest in a fund they probably knew full well had terrible risk practices and was likely going under on the first drawdown.

When Leopold went to pitch NY investors they all passed and thought he was full of it. He could only convince California tech guys. Savvy finance guys saw SA for what it was (leveraged beta trade). Jane street are finance guys, not California tech bros.

One theory (I don’t necessarily believe it): Jane Street was after the private part of SA’s portfolio (Anthropic), which is difficult to come by. Losing a few million dollars investing and getting in that network was worth it for them to try to groom Leopold to eventually sell them the private stake

It's all so sketchy. Jane Street were investors in SA but presumably were much more sophisticated and savvy than Leopold. When SA got in trouble, 3 firms got into a bid war for the assets at fire sale prices: Citadel, Jane Street and a third I forgot. Citadel outbid the other 2, but it's all weird, like Jane Street wanted in on the popular boy's book that they knew was going to tank and just were waiting around in the water like sharks.

Agreed, it's that the free market _allows_ you to do. What's concerning is that we're used to the "semi-benevolent" image of the free market where a company succeeds and then returns value to stock holders, and the employees and the investors and the country all share of the riches. Of course the executives and main shareholders get outsized returns, inequality might raise, but many people see a bit, and capitalism works, kind of.

This new Chinese "involutioned" capitalism is much more scary and robotic. The companies never really make money, becoming just an operation that maintains itself on minimum wage workers, all to the service of reducing prices. Shareholders don't make money, employees are semi-poor, nobody sees any returns.

But the customers see cheaper prices! you say. Sure, yes, the products get cheaper, but now think of yourself as an employee not a customer. Imagine the involutioned companies triumph. Your previous company, where you had a good salary goes bust. You find a job in an involutioned company where everyone is almost poor and worked to death just to reduce prices. Your portfolio goes to shit because companies don't make money anymore.... Sure, everything's cheaper now but I'm not sure that makes it better.

re: the crisis of the Mittelstand in competing VS China, I'm reminded of this tweet. It's a fictitious illustrative case not a serious study but I think it describes some of the root challenges of western industrial culture when it comes to competing against China

https://x.com/AngelicaOung/status/1921364873444986888

Basically yes, there will be more coding in that phase, more prototyping, the PRD phases will be shorter too, there will be more pressure to deliver quickly and the PRDs will be under more pressure to move more quickly. This is what I'm already seeing to be honest.

Coding was expensive in the sense that once you decided what to do, it took a few engineers months / years to do moderately complex projects. That's not true anymore. Therefore the risk of "coding the wrong thing" is less.

Like many others in the comments, I feel there are a lot of assumptions in this piece. Before, coding is only 14% therefore, small slice. I think that's a very superficial assumption. That was because coding was expensive and we needed to be sure we didn't code the wrong thing. If code is as cheap as it is now, we will optimize differently, we will structure around it. Instead of so many meetings we will code 5 different versions of the same thing and choose, etc.

I mean, so many people also went in on the boom, that's why it's a boom. Leopold somehow got it 100% right and _still_ managed to go bust

I think the collapse of SA is very simple (as long as I'm not wrong about it of course hehe).

SA weren't geniuses, they weren't sophisticated. They just did the same thing everyone else did, all in on semi-conductor, AI and memory positions. They got great returns because 1) everyone got great returns and 2) they were leveraged through their ears. In fact, not only they weren't geniuses, they were pretty bad at risk management, so bad that the first mild drawdown triggered margin calls on their over-leveraged bets and they couldn't cover them.

TLDR: SA didn't have alpha, they just looked good through over-leveraged beta and got caught

Thanks, that's insightful. I haven't thought about it that way but in retrospect it's clear. In practice I think we've all seen that the numpy style construction is quick to write and performs (and reads) much better than "dumb loops", but if you really want to optimize your code, the highway version will give you more control (and will read similar to the dumb loop with some decorations). I guess just more tools in the toolbox!

Sure, but we're not talking about most code, we're talking about "code that would benefit from SIMD"

Isn't the better abstraction here to use a higher level library in the style of pandas/polars that will operate as vectors, compose and feel readable and inuitive, while (almost?) maxing out SIMD?

When my daughter was 7yo and she talked to chatgpt, she repeatedly asked, what's your name? are you a boy or a girl? eventually she lost interest at the evasiveness of the answers, like she marked it as not fun to interact with, not a real person. It's different now, that she understands it can help with homework :D

I have this tshirt. I'm relieved it's not a cognitohazard, virus or something, thanks!

I used to play EVE on and off for years (decades?) and as many people here I loved more the idea of the game than the reality of it so I couldn't handle it for long periods.

I would love the possibility of setting EVE servers where we have like instanced pre-loaded scenarios (battle for a structure for 50v50 preloaded fits, alliance tournament style fights, etc.) and can do open source. Of course they'd need to open source quite a few more things, I wonder if they could franchise the game itself this way so people setting the servers pay a bit or something (or rent the capacity, cloud style?).

AFAIK, EVE is very profitable and EVE's mother company have been trying to use this cushion to branch out into other games and exploit the IP. I believe this has been mostly failed attempts and they've been bought, sold, etc. Maybe franchising EVE itself would work better? CCP plz?

I don’t want to be a conspiracy theorist but how much do we think this responds to the Trump admin having documented financial ties to openAI? Trump has proven to me very transactional in dealing with private business.

FWIW I agree the financials are a bit crazy and OpenAI went a bit nuts with the circular deals. That said, honestly, I don't think it's the end of the world. I think there will probably be some correction/crash and it will probably be healthy. A lot of these circular deals will get canceled, but it's at the end of the day people changing imaginary numbers with each other. The underlying tech I still think it's revolutionary regardless, the same way that the internet was and the tech boom crash at the end of the day was a distraction from the fact that these companies did end up "ruling the world"

There's no shortage of sources of information. I'll exercise "critical reading" with sources I consider trustworthy to begin with. I've no time to engage with difficult analysis from people who are not worth the effort. You wouldn't engage with every lunacy you read on a tabloid, right? similar principle

I guess what I'm saying is that I won't look at his numbers since he's an unreliable source from my point of view and there's a chance that he's going to try to deceive me and it's a waste of time for me to listen to him.

I do have other sources of information and I probably agree in general that AI companies are doing pretty shady financial shenanigans. I even think it's possible that openai is in real trouble. But I don't extrapolate that into "AI is useless", which is what he does.

Already many comments saying this but Ed Zitron is not a person I trust. He's been so biased and wrong on stuff that I consider very obvious and trivial that his complicated analysis with numbers and trends I can't just take at face value.

As an example of obvious wrong things, I remember a tweet of his where he was mocking people talking about agents and agentic coding. He was kind of saying that he was going crazy as agents weren't a thing really and people talking about them like they were real. Something like "agents?! what agents?! these guys hear themselves?!". The answers were full of hundreds of people patiently explaining how they were actually _using_ agents. This wasn't in 2023, it was a couple of months ago.

He just has an audience and an engagement target. His objective is clicks, not informing.

I don't think this works, and other comments have also pointed out the same. You open a position, you receive 200 CVs. Ok, then what, get them all to come in for a couple days work? A company that needs only one position can't possible handle that.

Now you're amazon/google. You have 200 open positions in a certain site/country. You receive 10000 CVs. Same problem, different scale.

So ok, you need to filter CVs, mmm, which sucks, right, so perhaps we do a screen interview? mmm, low signal, maybe....

Truth is, most people who interview people have no idea how to do it. I know because I've done hundreds and nobody ever trained me or explained to me how to do it properly. Over the years I've seen so many people on both sides of the table that I developed a method and I got semi-functional at it but so many people doing interviews shouldn't that bad experiences should be almost expected by now.

Even if we consider what they pay for colossus high (I've no idea, I haven't looked at the numbers), wouldn't this a bit be different from "investing in infrastructure"? They're not building the DC themselves, they're just renting, they can scale down to 0 anytime and not have pressure to recover costs, they don't have debt on the HW, etc.

Reminds me of the Ishkur's guide to electronic music

https://music.ishkur.com/

Both these maps of styles have most of their richness in the past. Modern era is mostly stagnation. I suppose it would be different if I had a map of hip-hop?

I feel there are like 50 load bearing assumptions in this piece of fiction that are dubious at best, definitely speculation. When you put all of them together and chain them in a product of probabilities it becomes tech bro wish fulfilling fantasy. But as usual this is expressed with total confidence and inevitability. This is a perfect encapsulation of all that is wrong with rationalists, TPOT, etc. IMHO.

I'll just suggest that you track your predictions and "update your priors" once they start domino falling.