I'm wondering how people like Musk, Zuckerberg, Altman, etc. just continually get away with it. Are they just more sociopathic and cover their tracks better?
I'm not really making a judgement on Huang, but it seems others have done worse. It's making me wonder if there's something larger at play where people (i.e., other billionaires and CEOs and perhaps politicians) are unhappy with Huang.
I suppose that's one reason. But Musk has formed basically pyramid schemes to keep his debtors from calling and to move his debt out from under him. Without that, he likely would have to sell shares to cover the debt of his Twitter purchase.
Musk has a history of having his companies buy his other companies, and his most recent example of this is obviously xAI buying X and then SpaceX buying xAI. For example, the purchase of X by xAI valued X at a ridiculous ~$40 billion. And then he took SpaceX public after all of this. And SpaceX is built upon $38 billion worth of taxpayer money, grants, tax credits, etc. Must be nice to have a bunch of loser companies all buy each other and then use a government funded hype stock to make it all back.
It is absolutely impossible to know the exact details, but Musk has a history of shady deals (even having his companies bill each other for expenses), and so it's not much of a leap of faith that these moves are to absolve himself from lenders and investors by shifting debt around and moving losers into winners.
I bet that SpaceX will buy Tesla any day now.
Musk somehow weasles his way through lawsuits and gray areas and corruption and legal loop holes to just have his companies buy each other out when they start sucking, all on the teat of others and the government.
They could lose control through a substantial shareholder lawsuit. A minority shareholder suit against Musk could prove tremendous damages repaid in stock and the judge might not let the special classes of shares with extra voting power protect his control.
It would have to be a very good cause though, disliking him or ordinary bad business decisions won't cut it.
Musk, Zuckerberg, and Altman and their companies have zero history of sketchy stuff that would, in a normal world, lead to CEOs being ousted? Affairs are enough to do it to Huang? The whole Chinese spy thing is not confirmed is it? And the three people I've listed have had documented cases of endangering national security through either their actions or their company's actions.
"Okay, when I introduce you, I'm gonna say, 'This is a friend of mine.' That means you're a connected guy. Now if I were to say, this is a friend of ours—that would mean you're a made guy. Capiche?"
I got this book today, as I had it on pre-order. It's a bit disappointing because, on a quick count, there are only five pages with pictures. And I think I saw just a few equations. It's a bit strange to leave out two essential mechanisms for how mathematicians work and think in a book named Six Math Essentials.
> Of course, at the moment everything’s still a bit very much difficult to tell, but “productivity”, at least in applicable areas of work, might really come down to who can let go of the most understanding the easiest? Incredibly thought-provoking stuff.
I cannot see it being a good thing to "let go" of understanding. If you don't understanding anything, then why would anyone keep you around? An example hypothetical situation is, if I have a colleague that outputs a bunch of productive things, but they don't understand it, and I do, then of what use are they? I have LLMs too and also have the understanding.
It isn't clear to me why people are so comfortable with letting go of understanding. And it's a bit bizarre because it always seems to be framed as "you're going to get left behind if you care about understanding". I also never hear or read anyone who has any real and deep experience say these things.
My read of what is being suggested is that the level at which one should aim to understand and be comfortable with is slightly more abstract than reading the code and reviewing every line of it. It is like how one is comfortable not understanding the byte-level code or assembly language code. I know that this analogy doesn’t hold very well because compilation is a deterministic process while llms are not, but that is precisely the point being made. “Trust” that llm will produce the right output, even though the output might not be reproducible word for word the next time you give the same prompt.
To your point about becoming redundant without understanding, not understanding things will be a problem when an issue arises and the other person isnt able to handle it the right way. However, if llm can address the issue, does the fact that they did not understand the details really matter? From the perspective of a person who gets to decide who gets fired, would they fire someone who is productive but sometimes introduces bugs that they quickly address or would they fire someone who is comparatively slow but rarely introduces bugs? The answer probably depends on the domain, and how costly a bug can be. For most of us, the work is not so critical that you cannot afford bugs. So, productivity wins.
If you are suggesting humans before LLMs rarely introduced bugs, you probably haven’t got the latest models to review your old code. It’s shocking how many bugs we had before but couldn’t see!
Don't shoot the messenger, but maybe the idea is this:
You're smart, and you know what can go wrong, so you use models slowly, and you check their work. Your bet is that it will make a bad mistake.
Your coworker is reckless, they don't know or care what can go wrong, so they use models quickly. Their bet is that the model won't make a bad mistake.
If the model doesn't make a bad mistake, your coworker is creating more value to your employer, and your employer doesn't really care if you understand the code. They just want something that works.
Over the last year, it sure seems like models are making less mistakes, so the expectations on the bets are changing too.
Right or wrong is this the future we want? Shouldn’t we say and do something?
For instance: we have calculators. We’ve had them for decades in their “current” form more or less. They are readily accessible and far more reliable than LLM’s - if you a problem 1000 times it’ll be right 1000 times, the rare bug or glitch notwithstanding. But we still teach people basic arithmetic. We teach times tables. We didn’t just offload basic math.
I agree we don’t need to know assembly to use a computer but there is a balance here and companies do not care about a balance.
I think we mostly DID offload basic math, for better or worse.
For software engineering I don't know the right balance.
I've used the "we mostly stopped caring about machine code generated by a compiler; maybe we should stop caring about LLM-generated code as well?" analogy myself.
But even I have to admit there's two big holes in that analogy: if I use a compiler to turn C into machine code... (a) compiler bugs are orders of magnitude rarer than LLM-generated bugs (b) my C code remains the source of truth, or at least source of intent. Whereas with LLM-generated vibecode, there's no source of truth, just an ever-evolving mudball. I (kinda?) like the ideas of "documentation as source of truth" and/or "tests as source of truth" ideas, but I don't think we're nearly at the point of viability.
TL;DR yeah I think actual understanding of what LLMs produce is gonna be important for the forseeable future
Like humans, LLMs also have to expand more effort when dealing with large piles of fragile spaghetti code. Spaghetti code isn't just aesthetically displeasing; there's a objective complexity increase and for large enough apps the increase is exponential. It's hard to imagine this ground truth ever changing.
Like humans, LLMs get into a place where fixing one thing breaks three others. They compensate partially with insanely big test suites, but now fixing one thing breaks three others, and 57 different specs, half of which also need to be updated.
I think roughly three outcomes are possible.
1. LLMs get fast/good/cheap enough that this just doesn't matter. Your 50M LOC monolith is a mess? Yeah, it still takes more effort for the LLM -- but this just means it costs 5 additional seconds and $0.03 instead of $0.02 to ship a new feature.
2. Stepping away from the "code as source of truth" paradigm. What if specs/docs were the vetted source of truth, and we just regenerate the whole 50M LOC monolith every time somebody makes a change, just like we rebuild binaries from C today whenever there's a change? (This probably requires some of the LLM advances from #1. Also, at this point, perhaps we should just skip the intermediate language and generate ASM -- what purpose is human readable source code serving at that point?)
3. Status quo as I see it -- human engineers overseeing projects, both guiding AI and being accelerated by it.
I think #2 is probably the... most ideal future? Though, of course, if we switch to that paradigm, the source-of-truth docs will have to be very thorough and structured, and will probably resemble a programming language of sorts themselves, except describing intents instead of operations.
Adults can, under most circumstances, do some limited slow arithmetic but are generally piss-poor at it, notwithstanding the occasional rainman, and only rely on it in trivial matters.
You see cashiers summing groceries? Salary calculations on paper? Tax, inventory, accounting, you name it? Nobody does any of it by hand.
I didn’t say we were universally good at it but in many countries we are (more or less) all exposed to it and taught it in school. There is a baseline even if we are far from mastery, and frankly a cynical hand wave of “well everyone’s bad at it” is not a real argument. In the US it is very clear that people are at least exposed to it and if we weren’t we would be operating very differently. I went to dinner last night, everyone did the tips in their head right there at the table. The average person can do some simple multiplication and division in their head. It’s not a mysterious script done on a box that they repeat the output of. Most people even if they’re bad at it understand the fundamental concept and, gun to their head, could probably do a series of very basic math problems with pencil and paper.
I imagine you and most of your peers can calculate a tip or at least approximate it without pulling out their phones yes?
Analogies can be good when you're explaining something you know, but they kinda fall apart when making arguments or decisions.
How many of your friends and extended family can write hello-world or fizzbuzz in any programming language at all? Should they be allowed to vibe code apps for themselves? If they get good at vibe coding, should an employer be allowed to pay them for it?
> Analogies can be good when you're explaining something you know, but they kinda fall apart when making arguments or decisions.
If all you’re going to do is just hand wave away anything I say then this isn’t going to be a productive conversation. Have a good rest of your week man
I was trying not to nitpick the ways arithmetic is different from programming. That wouldn't have been productive either. I wish more people were good at math, I wish more people learned to program, and I hope you have a good week.
This is just moving to a higher level of abstraction. Nowadays people don't verify the compiler generated assembly code anymore. We don't care if our source generates a few million extra instructions. Source code is just an intermediate between specs and binary at an abstraction level decided by the developer. We do requirements -> code -> assembly.
To resolve this we should aim for a new intermediate format that is more high level than source code, the generated code is too monotonic to properly review manually. Something like requirements -> detailed specs for llm -> code -> assembly . Then let one agent implement the specs and have another review it. If we stay at the current abstraction level, humans become the major bottleneck.
> Nowadays people don't verify the compiler generated assembly code anymore. We don't care if our source generates a few million extra instructions.
This analogy doesn't hold up. Compilers are designed programs and are deterministic. We "know" what they generate, and what they generate and how they generate it is well-designed and tested, and there are certainly people and teams who know how the compilers work.
None of that is true for LLM-generated code.
> To resolve this we should aim for a new intermediate format that is more high level than source code
That was the problem before and after. LLMs did nothing to address that. This is what the entire software engineering domain is about: translating loose system designs and requirements into programs.
LLMs do not replace the code, they replace the coder. The coder becomes the manager, and the model becomes the coder. The code it writes is ultimately deterministic and can be tested. The LLM coder may still hallucinate, although that can be said of human coders too. In the classic development model, we use specs and tickets to steer the output of a human coder. My take is that the same needs to be done for LLM coders.
I think your hypothetical is incomplete.
What if the person that lets go, generates more productive things then somebody understands?
In this case I know a lot of ppl will value the producer of value more.
I always pushed the bounds of learning more about the underlying stack or problem, and always found them in no time.
Understanding architecture, robustness, the “crumple zones” of a system? I’m all for it.
Taking three days to “understand” how postgres transactions work under the hood for your generic SaaS? Uh..
I have seen a lot(!) of the last type of “understanding” and it’s infuriating. The same types will often not understand the business, its stakeholders, the architectural constraints beyond the technical, etc. Understanding takes many forms.
I feel like the real value is in probing the extents of your own understanding. If you don't understand something it's doing, you can ask. You do kind of move up a level of abstraction, from the IC to a PM or team lead. So maybe you don't know what every line of code does but if you understand code and how to drive you can trust that it's doing the right thing and you don't have to be as deep in the weeds of every refactor etc.
You do have to keep a sense of when it did the wrong thing and how that was a valid interpretation of what you asked and how to better specify what you want. But that's how it really just shows you the gaps in your own understanding of a problem.
Also I think there's huge differences to the kinds of tasks and workloads people are throwing at it, leading to wildly different perspectives, but everyone's talking about the same "Fable".
> Also I think there's huge differences to the kinds of tasks and workloads people are throwing at it, leading to wildly different perspectives, but everyone's talking about the same "Fable".
This right here. Humans are really prone to over indexing on a handful of personal experiences vs a representative sample. And once someone has formed an opinion, confirmation bias helps cement it, despite evidence to the contrary. Reality is probably somewhere in the middle.
That's where I think benchmarks are a huge red herring. Everyone wants a simple performance evaluation of a repetitive task but that's missing the point entirely IMO. Just try talking to it, asking questions. I spend so much time just reading responses. I have to like really think about how to specify the question I'm asking, fire it off, and I often can't wait to see what it says and what I'll learn. How anyone thinks it's making us dumber is crazy to me...you're missing out if you're just trying to offload thinking.
You're misinterpreting what they mean as understanding and inserting your own doomer definition where this imagined person has been stripped of all ability to investigate a result or ask questions.
I too love to downvote and edit instead of admitting that I'm talking and acting in a way where nobody would want to be my coworker anyways. Where exactly are you demonstrating that you're doing anything to justify being kept around if that's such a big concern for you?
There are many ambitious things I have tried to work with Opus 4.8 and Opus 5.5 on, and it just can't get there. It always falls back to some tried and true method or some valley that it gets stuck in. In those times, it was impossible to push it forward into more creative and exploratory thinking based upon my relatively advanced pushes and contextualizing. So I took the commenter's position that Fable 5.2 will respond better to that type of pushing. But that type of pushing actually means you know more than those who are happy enough using the lower models. So that seems like the opposite of what you said.
I'm closer to a Product Manager so I'm not sure how useful these examples will be for you.
One important constraint is this topic of intent (also referred to as alignment). Specifically, when a user requests a feature or change, there is MOUNTAINS of implied context and intent in their comment. Some cultural, religious, social, economic, base on work and personal experience, their use cases, their values, what they intend to get done, etc. Every time these models get smarter, they get a little better at understanding this context. It's combination of reasons. Some of it is parametric. Some is inferred. Some is about selecting the right data to examine and research. As these models get smarter, I spend less time time iterating and educating the model. They reach the intent faster and better.
From what I hear and experience, this is also true in the coding space. LLMs are basically another abstraction level on top of formal languages, which sit on top of compilers and machine code and etched instructions. Smaller and dumber models require a lot more hand holding to achieve the same results. They're less capable of inferring the gaps. Less capable of extrapolating potential negative outcomes of the change they're making. Less/less useful parametric knowledge. Less capable of selecting the right data to use to make the right decisions. They do what they're told.
This has been less of an issue for coders who religiously document, understand the codebase and architecture intimately, and demand to be kept in the loop on every change. So seniors often do not see the utility of smarter models. They know exactly what they want, how to ask, the context to provide, and how to assess if the change worked as expected. The utility is for the 99.9999% of the world which is not a senior developer.
While what you say mostly rings true to me, I don't see how it could result in the valleys of stuckness I was asking about.
However, your statement that there are only 8300 senior developers in the world does not seem correct to me. That seems like an unreasonably high bar for that title.
They are generally things I don't particularly want to share, as I view them as ideas close to my heart. The general pattern was what felt like redirecting from the harder problems and an unwillingness, if we can call it that, to go there. In some sense though, it's hard to understand how these more advanced models get around that, because these models are tuned to the most probable answer. When you come at them with a novel idea, they always seem to water it down in the end.
I have also found this — once you get out into some really novel workspaces they will revert back to replication of adjacencies that already exist. Fable 5 and 5.1 in particular will talk around the issue in a way that no model prior seemed to.
I think the issue is fairly fundamental, perhaps because of tuning, perhaps even beyond what they are tuned or trained on, because the bulk of what they can go find and ground into seems to drag them heavily toward implicit biases that split around categories I can’t quite discern. The category boundaries sometimes look like academic fields, but sometimes they seem like more fundamental attractors.
I second (third?) this. Previous models were more willing to explore the unknown. Those new ones just try to Google search and latch onto any adjacent solution they find.
It's a reference to a joke, of a stereotypical comment left on Romantic Relationships topic online forums. As in, "this man is so terribly unsuitable to everyone. But I'm special, and I can fix him and make him the perfect boyfriend."
The comment is implying that the commenter thinks that you think that with more information you could solve the problem, and that is why you asked, instead of trusting the assessment/competence of the other person.
I am not sure you thought that, you just seemed curious, but that is the relevance.
The aphorism I used there, if you substitute both instances of the word language with war, was coined by the article author in their most famous novel, Blood Meridian.
Read the original comment I replied to. You suggested that Huang has leverage in this situation. I suggested otherwise because I'm certain he will be both financially and emotionally keen to stay at the helm of the company he has helped build over the previous three decades. So no, I don't think he has leverage because the board knows that he's financially and emotionally keen.
Now you and the other commenter are suggesting that he could just go off and start a new competitor? For one, literally no one has done that, and the only competitors are already massive semiconductor companies. Secondly, his "assets", as you put them, are almost entirely tied up in NVIDIA. So you can fund a competitor using stock of the company you hope to beat? That makes no sense. Thirdly, I highly doubt a separation agreement, if it came to that, would allow him to start a competitor.
> I'm certain he will be both financially and emotionally
Financially, no. Most of his stocks are already vested. Emotion is just a soft condition. It's not like he would go bankrupt without a job.
> For one, literally no one has done that
Tons of people were fired from their own startups and used the money to start, fund, or work for a competitor.
> Secondly, his "assets", as you put them, are almost entirely tied up in NVIDIA.
He could sell NVIDIA stocks... especially after he was fired.
> So you can fund a competitor using stock of the company you hope to beat?
It's so naive to think one company will beat another company.
Jensen would be the best candidate to start a chip company that could become a major player.
Semiconductor companies today are pretty happy because there are like 3-4 players. Having another player coming into the industry will be extremely expensive for the current players. Nobody would beat anybody. But the market share can be shared with more players.
> Thirdly, I highly doubt a separation agreement, if it came to that, would allow him to start a competitor.
They cannot force Jensen to sign that term. Sure, if Jensen accepts that term, it means he doesn't want to start a competitor. But we don't know what Jensen is thinking.
We are talking about the possibilities. Starting a competitor is one major possibility that the board has to consider if they think about firing Jensen.
> So again, I ask, how would that work?
Are you asking how to start a company e.g. setting up LLC in Delaware? or you are saying it's impossible for Jensen to start a new company after being fired from NVIDIA if he is fired?
I'm not a billionaire, so I don't know whether it is difficult for a billionaire to start a company. Maybe this is one thing that is easier for average people. I have no idea.
> day 1 security best practices say you don’t write secrets to files
If only tools, including major ones, actually did that. Plenty of the biggest package managers in the world still write API keys in plain text to files. uv and npm are two of them.
> Football was most definitely a net-negative for my life. All I learned is how to be homophobic, how to be hateful, and how to get hurt.
I played football and yet never drink, have never done drugs, am not and was never homophobic, never got into trouble etc.
Would I let a kid play football now, with all the knowledge of CTE? No. I absolutely even worry about it for myself. But it's a bit odd, to me, to say "football made me a bad person".
I mean I’ve of course unlearned the regressive social circumstances of football, nor is it unique to football. It’s fairly common in many male-dominated spaces, particularly ones which tend to cater to toxic ideals of masculinity or, as it is sometimes euphemized, comradery.
I think basically all people who were socialized male have had these experiences at one point or another, whether it be in football, in class, in the military, at sleepovers, in the locker room.
My point is more so that football has a particular “culture”, or it did when I was a kid. And that culture was not necessarily socially productive at a young and impressionable age. When we see certain behaviors in particularly young men, we have to wonder where they come from, and what we are doing to cultivate that.
I'm not really making a judgement on Huang, but it seems others have done worse. It's making me wonder if there's something larger at play where people (i.e., other billionaires and CEOs and perhaps politicians) are unhappy with Huang.
reply