I think the general argument is “if you can’t bother writing a blog post, then why should I read it?
Which seems like a pretty good argument to me, given the amount of slip out there and how increasingly performative these pieces are.
1,193 karma · joined December 13, 2022
Which seems like a pretty good argument to me, given the amount of slip out there and how increasingly performative these pieces are.
A waste of time. Every. Single. Time.
It wasn’t quite as needlessly performative. But it’s never been good.
It’s just not particularly common (or helpful) to view them that way.
Long term, this only works if you have a non-commodity, and if the higher tier is actually more profitable. We'll eventually learn whether both are true. For OpenAI right now, it's probably enough to just increase revenue, even if the higher tier is even less profitable.
You have some startup, the founder is either young or doesn't want to do the CEO stuff. Things kind of eek along until the founder either steps aside or is removed by the board because a) it's time for a "grown-up" CEO; b) the CEO needs real sales experience; c) the founder overpromised and under-delivered; d) board/VC politics make it helpful to install a buddy as CEO; e) etc.
Then the CEO clock starts, typically they have 18 months to get lucky and hit their metrics. They do a lot of glad-handing. They hire "their team" of sales/marketing/etc people. They spend A LOT of money. And I mean A LOT. They talk about OKRs or SMRTs and KPIs. Out of nowhere a small army of project managers show up and try to tell you how to do your job and why you can't just talk to the <thing X> team directly but have to go through them for "efficiency" and "visibility."
In 3-6 months, senior engineering and R&D staff starts to leave. Whatever culture you had slips away. HR has "culture" meetings to "find the right company culture."
Sales/product can't sell and points the finger at R&D, maybe even for the right reasons. You OKR/KPI harder, but it doesn't matter because nothing addresses whatever the underlying problem is. Multiple senior people have pointed loudly to the problem and are ignored; they're often not managers so it's unclear if they were even heard.
At some point there are one or more rounds of layoffs; sometimes these are announced, sometimes it's just a gradual attrition.
Eventually the CEO clock runs out. They don't get lucky. Nothing they did helped, and some of it hurt. They collect their $1M severance, get to keep their stock, get 9-12 months of health insurance, and move on. In a year or two you hear about them joining a new company as CEO.
In the meantime, you've either moved on or have a new CEO with a new 18-month clock.
Almost like the Star Trek mechanisms can infer perfect intent.
Like there’s a hidden script or something.
More seriously, I think there’s real value in an automatic door that behaves consistently rather than one that tries to infer messy human intent. Real life isn’t a TV show and there’s both ambiguity in how people behave and how they even intend to behave. It’s mostly not hard to understand how a proximity sensor door will function. Using a black-box classifier to improve that won’t necessarily make people like it more. And calling up to the cloud for every sensor event, ignoring privacy issues, adds weird latency and a huge failure mode during data center outages.
Being unemployed is hard and can be really demoralizing. Try thinking “I’m probably helping someone pay rent until they find a job listing that works for them” when you’re rejecting obviously unsuitable applications. It at least makes the “wtf”-ness of some of them seem a little less weird.
This is wrong. Reporters frequently don’t understand the science or the nuance in the science.
Reporting and science are two very different disciplines. Reporters rarely have a deep background in science and almost never have a background in the specific area that they’re reporting on.
Hell, even scientists have trouble accurately describing the work of a different scientific discipline.
Don’t invent bad faith motivations; they exist but most of the time it’s just two people slightly talking past each other.
If the OS knows that a thread is waiting for a lock, the scheduler will not bother to schedule it until the lock is available.
In general, it’s tempting when you’re bound by lock latency to skip the syscall overhead of sleeping. But a lot of the time that’s a code smell that there are other inefficiencies in the system and you should rethink how you’re scheduling work.
The dystopian part of me suspects that this is just a temporary blip as Amazon pivots from ads designed for humans to ads designed for agents. Which, in its truest form is just Amazon being a middle-man for bribing or poisoning agents.
In the end, it’s hard to not view everything through a lens of back dealing and anticonsumer enshitification.
Or maybe I’ve watched just one too many influencer videos on how the AI bots are either going to destroy humanity or take all the jobs.
It’s so common that there’s an xkcd about it:
The job market sucks, but there are so many confounding factors: pandemic over hiring, the loss of ZIRP, an economy that would be recession if it weren’t for data center investment, and may be in recession soon even with it.
This industry has been in down cycles before. It’s just been a while. So far it feels like a down cycle, not the sky falling.
It’s not clear what the future will hold, but I’m 100% sure it won’t evolve the way Anthropic predicts. Both because they have strong biases and because there's just no way to predict how this technology will play out. Humans are, if anything, remarkably adaptable.
I was with you up to “usually by coincidence”.
There’s a long and sordid history in areas of chemistry and areas of biology of holding up a competing paper in review so you can scoop them. I’m sure it exists in physics as well. Certainly biophysics, but probably most subfields.
Often it’s a famous labs that can steamroll review or even just dump the work into PNAS as a “member contribution.”
At least one author of a famous inorganic chemistry textbook was rumored to do this routinely.
And I know of at least one National Academy member who swore off arxiv prepublication after getting scooped.
None of this makes it all right. But plagiarism and academic theft is old and definitely not always accidental.
In a stable, profitable company, you need sales to keep the business solvent. And understanding the sales pipeline is pretty essential to effectively running a business.
Where this breaks down is when the C-suite forgets that you also need a product, and you need to keep innovating on that product in ways that bring value to current and future customers. This seems to happen pretty frequently.
It’s true that collective groups of humans can accomplish tremendous things, but I hesitate to use “superhuman intelligence” for corporations, governments, and institutions.
So much of what they do ends up being very focused on process for the sake of process. And is often so very dumb.
You raise some good points, but I don’t think the idea of moats from the 80s really holds up. None of the obvious moats in the mid-80s are still dominant players. In the mid-80s the PC market was still pretty new. Business was still run on IBM mainframes, DEC VAX, and other systems that were overtaken by forces in the late-80s and early 90s. These seemed unstoppable until they weren’t.
I think the strongest counter example, though, is probably Linux. If your thesis was correct, an early-90s hobby operating system by some kid at a uni in Finland would never be dominant in cloud server spaces by the 2000s.
> Amiga couldn't over come the PC.
This is probably your best example. I would add BeOS to this.
On the other hand, Apple managed to attract a core audience and stay alive (even if barely) through the 90s. Sometimes you don’t have to overtake; you just need to have a solid niche.
> IBM couldn't over come the moat of Windows with OS2.
OS/2 originally lost to Windows as a successor to DOS due to execution, not because Windows had a huge moat. OS/2 Warp was a bit later.
The story of OS/2 failing is sad and as much about IBM internal politics as about Microsoft succeeding.
> Intel couldn't overcome x86 with the i432 or Itanium.
Intel bet on compilers being far smarter than they were in the 2000s (and possibly today) and didn’t seem to understand that most business applications at the time were limited by branching and integer computations. That stuff doesn’t vectorize well. The arch might be better-received today, although moving pipeline concerns to the compiler still seems like a bad choice to me.
As a counterpoint, consider the rise of ARM in the 2010s. An arch known mostly in the embedded world has become a real player in the desktop and server environment. All because of the rise of phones, reasonable licensing, and better perf/watt at low power than x86.
Perhaps that’s true, but it’s a price tag that’s a lot harder to swallow for many orgs than $200/mo, and would require some hard justification for how your increased productivity contributes to the business bottom line.
I’ll agree that you can probably do that with hard numbers. I am skeptical that most $200/mo users could.
Turns out when you have the choice of accepting lower payment per patient for Medicare or having a lot fewer patients, you choose the lower payment per patient.
I would expect the same situation here. Doctors would grumble, but no one would force them to accept patients on whatever “Medicare for all” would be called. Nothing other than market forces.
A number of things would probably have to change, including the cost of medical school. But the system right now is expensive and essentially unsustainable. So change is inevitable.
Actually, I spent a considerable amount of time in my doctorate and postdoc doing this.
Any kind of MCMC sampling of a simple model tends to be bound by the rate you can draw variates.
Examples of this include: Gillespie simulations of chemical kinetics, Ising and Potts lattice models (including their roughly bazillion variations), and anything resembling bootstrap or permutation sampling.
Just because your problems aren’t bound by the rate of drawing uniform variates doesn’t mean that these problems don’t exist. It just means that you have a narrow view.
As someone who has spent considerable time working in these areas, I still appreciate advances.
It’s a cute way of saying that you can only do 2-3 new things.
The post is written for an engineer at a startup as a reminder that although it’s green field development, you only have so much runway, so it’s better to focus on what matters instead of trying some new tech because it seems cool.
If you’ve ever had to estimate your stories/tickets/etc in “story points” or “T-shirt sizes” then “innovation tokens” is roughly the same.
If you haven’t had to do that, you’ve lived a charmed life.
> Engineers should understand requirements, risks, tradeoffs, and potential gains.
Ideally. But I’ve worked with plenty of engineers who get far too excited by shiny new tech and overvalue its potential while undervaluing its risk.
Hell, I’ve been that engineer in my misspent youth. The post resonates with many of us because it describes hard-won wisdom of our mistakes.
> New technology may be right for that. Novel approaches may be right for that. "Novel" or "New" are only proxies and they're weak.
Yeah maybe, but unless you’re working on a problem that the tech directly solves, it’s pretty unlikely.
> What if you know NodeJS really well? Or MongoDb? What if you have empirical, verifiable reasons for why they fit better?
In 2015, MongoDB was a dumpster fire (which is still kind of true) and node.js was still kind of new and had enough rough edges that most teams were probably better off choosing some other language/framework.
> I'm a bit tired of "simple" and "boring" and other nonsense words in this field taking up the air in the room that should be spent evaluating solutions on their actual merits.
“Boring” and “simple” are ways to convey that it’s good to be risk averse. It’s a bit of rhetorical flourish that helps drive the point home: choose what you work on carefully because you have limited runway and should spend that runway working on the problems that matter for your business, not new tech that’s orthogonal to it.
But the core question of “how does this behave as N -> \infty?” is asymptotic behavior (ie: limits) which were developed for calculus and are very much part of the foundational calculus canon.
I loved graphing calculators until I learned tools like Mathematica and Matlab. Still waiting for the Mathematica version of LLMs.
Agents / loop engineering / whatever is hot with the AI Twitter kids still isn’t it.
Given that the term is explicitly mentioned in the Chicago Manual of Style, I suspect that this is much more of unicode trying to capture existing behavior, even if rare. And I think it's important to err on the side of capturing what people did, even if it's from an bygone era.
> How many people actually use this thing? 10 loud people with power using it does not justify putting it into the open standard.
It's definitely more than 10. But even if it's 10, I disagree with your assessment. People used this; it's worth a codepoint. There are far stranger code points out there.
And I've seen far weirder things in open standards. Have you read the C99 standard? Are you aware of the three allowed integer representations? Of which maybe two were in use in 1999, and one only in use in relic hardware that might exist in maybe tens of sites.
As other people have commented, it's partly a relic of an era where things were typeset by hand. Some journals may also require this in their style guide; I don't know of any offhand but the list of journals is long and their choice of style is varied.
> I dunno, do people actually read through a citation list, in a manner where the redundancy would be relevant?
Yes, they do. When you work on a Ph.D.-level dissertation, you end up having to learn your subfield pretty well: its major works, open questions, current major players, their ongoing arguments, etc. You start to notice citations cluster, and you can often find more than one opinion group. People you have to listen to (ie: your advisor / committee) will hand you articles with authors you don't know and who may come from labs that you don't know. One common way to start orienting yourself is to look at who they cite.
You pick up patterns: some people love to cite themselves and their friends, some people really try to give a detailed overview of the state of the field, and some people go out of their way to attack their rivals via citation. Sometimes the reference list has an author or work you don't recognize, and might have missed while glancing over the text. Mostly it's something mildly relevant and occasionally it's an exciting direction to look into.
The really long citation lists, where the triple emdash would make sense, mostly occur in fields outside science and engineering where knowledge is exchanged through chapters and books. Some fields expect a lot of detailed textual analysis which can involve a lot of careful citation of the same chapter or book over and over again.
While I've never worked in these fields, I've been friends and roommates with people in them, and I quickly learned that bibliographies are way more of a thing for them than I ever imagined.
You’ve clearly never had to deal with a lot of references. Author lists can cluster. If you’re reading, it can get tedious if the last 5+ references are to the same three authors. Let alone 20+ references.
This happens. In some writing it happens _a lot_.
It’s a common enough that it’s even explicitly mentioned in the Chicago Manual [0], one of the main style guides for academic writing in the US:
> Em dashes also substitute for something missing. For example, in a bibliographic list, rather than repeating the same author over and over again, three consecutive em dashes (also known as a 3-em dash) stand in for the author’s name.
[0]: https://www.chicagomanualofstyle.org/qanda/data/faq/topics/H...
You can do a full unpacking-via-lookup with a uint16[256] and then do bit shifting and masking to extract the individual trits, but using an extra byte in each entry (or 3 tables) would let you extract with just two shifts.
This starts to vary a lot with the microarchitecture, and there’s the added dimension of SIMD vectorization, so accurate timing in a realistic context becomes important.
Even if the current President was known for his strategic global thinking and ability to keep quiet about the “real goals,” other explanations seem more likely.