This post has an unnecessarily aggressive style but has some very good points about the technology hype cycle we're in. Companies are desperate to use "AI" for the sake of using it, and that's likely not a good thing.
I remember ~6 years ago wondering if I was going to be able to remain relevant as a software engineer if I didn't learn about neural networks and get good with TensorFlow. Everyone seemed to be trying to learn this skill at the same time and every app was jamming in some ML-powered feature. I'm glad I skipped that hype train, turns out only a minority of programmers really need to do that stuff and the rest of us can keep on doing what we were doing before. In the same way, I think LLMs are massively powerful but also not something we all need to jump into so breathlessly.
I empathize with it, but ultimately it's fruitless. This happens with every big tech hype. They very much want people to keep talking about it. It's part of the marketing, and tech puts a lotta money into marketing.
But that's all it is, hype. It'll die down like web3, Big Data, cloud, mobile, etc. It'll probably help out some tooling but it's not taking our jobs for decades (it will inevitably cost some jobs from executives who don't know better and ignore their talent, though. The truly sad part).
> It'll die down like web3, Big Data, cloud, mobile, etc
At least half of those the promise was realised though - mobile is substantially bigger than the market for computers and cloud turned out to be pretty amazing. AWS is not necessarily cost effective but it is everywhere and turned out to be a massive deal.
Big Data and AI are largely overlapping, so that is still to play. Only web3 hasn't had a big win - assuming web3 means a serious online use case for crypto.
"Die down" in this context means that the hype will come, go and then turn out to be mostly correct 10 years later. That was largely what happened in the first internet boom - everyone could see where it was going, the first wave of enthusiasm was just early. I don't think any technology exists right now that will take my job, but I doubt that job will exist in 20 years because it looks like AI will be doing it. There are a lot of hardware generations still to land.
"The proportion of global companies planning to increase spending on AI over the next 12 months has slipped to 63% from 93% a year earlier, according to a recent survey of 2,500 business leaders by software company Lucidworks Inc. Meanwhile, just 5% of companies in the US are using AI, according to the Census Bureau."
To a first approximation, I expect companies to spend nothing on AI and get put out of business if they are in a sector where AI does well. Over the medium-long term the disruption looks so intense that it'll be cheaper to rebuild processes from the ground up than graft AI onto existing businesses.
> At least half of those the promise was realised though
I dunno, I think there might be different sets of "promises" here.
For example, "cloud infrastructure" is now a real thing which is useful to some people, so one could claim that "the promise of cloud infrastructure" was fulfilled.
However that's not really the same promises as when consultants preached that a company needed to be Ready For The Cloud, or when marketing was in a slapping "Cloud" onto existing product marketing, or unnecessary/failed attempts to rewrite core business logic into AWS lambda functions, etc.
AI and 'Big Data' (as trends) aren't really overlapping in my view. Of course training these LLM models requires a huge amount of data but that's very different from the prospect of spinning up a Spark cluster and writing really badly performing Python code to process something that could have easily been done in a reasonable time anyway on a decent workstation with 128gb of RAM and a large hard drive/SSD, which was a large part of what the hype train was a few years ago.
At a point in time the database was a bleeding edge technology.
Ingres (Postgres)... (the ofspring of Stonebreaker), Oracle, ... Db2? MSSQL? (Heavily used but not common)... So many failed DB's along the way, people keep trying to make "new ones" and they seem to fade off.
When was the last time you heard someone starting a new project with Mongo, or Hadoop? Postgres and Maria are the go to for a reason.
There's a team at my company that chose Mongo for a new data transform project about a year ago. They didn't create a schema for their output (haven't to this day) and I'm convinced they chose it purely because they could just not handle any edge cases and hope nobody would notice until it was already deployed, which is what happened. For example maybe one in a thousand of the records are just error messages - like they were calling an API and got rate limited or a 404 or whatever and just took that response and shoved it into the collection with everything else.
In the last few years I have come to think of AI as transformative in the same way as relational databases. Yes, right now there's a lot of fad noise around AI. That will fade. And not everyone in IT will be swimming in AI. Just like not everyone today is neck deep in databases. But databases are still pretty fundamental to a lot of occupations.
Front-end web devs might not write SQL all day, but they probably won't get very far without some comprehension. I see AI/ML becoming something as common. Maybe you need to know some outline of what gradient descent is. Maybe you just need some understanding of prompt engineering. But a reasonable grasp of the priciples is still going to be useful to a lot of people after all the hype moves to other topics.
I agree that the world isn't changing tomorrow like so much of the hype makes it out to be. I think I disagree that engineers can skip this hype train. I think it's like the internet - it will be utterly fundamental to the future of software, but it will take a decade plus for it to be truly integrated everywhere. But I think many companies will be utterly replaced if they don't adapt to the LLM world. Engineers likewise.
Worth noting that I don't think you need to train the models or even touch the PyTorch level, but you do need to understand how LLMs work and learn how (if?) they can be applied to what you work on. There are big swaths of technology that are becoming obsolete with generative AI (most obviously/immediately in the visual creation and editing space) and IMO AI is going to continue to eat more and more domains over time.
I’ve been doing just fine ignoring AI altogether and focusing on my thing. I only have one life. Fridman had a guy on his podcast a while ago, I don’t remember his name, but he studies human languages, and the way he put it was the best summary of the actual capabilities I’ve heard so far. Very refreshing.
> I work on all aspects of human language: the words, the structures, across lots of languages. I mostly works [sic] with behavioral data: what people say, and how they perform in simple experiments.
(I find it ironic to see a grammatical error in his bio. Probably because of a mass find/replace from "He" to "I" but still...)
He is, very clearly venting into an open mic. He starts with his bonfides (a Masters, he's built the tools not just been an API user). He adds more through out the article (talking about peers).
His rants are backed by "anecdotes"... I can smell the "consulting" business oozing off them. He cant really lay it out, just speak in generalities... And where he can his concrete examples and data are on point.
I dont know when angry became socially unacceptable in any form. But he is just that. He might have a right to be. You might have the right to be as well in light of the NONSENSE our industry is experiencing.
Maybe its time to let the AI hate flow though you...
As someone who spent an inordinate amount of time trying very hard to be less angry despite having a lot of good reasons to be, a chunk of which overlap with this piece, I get a lot of dismissal from people who seem to think any expression of any negative emotion, especially anger, deeply discredits a person on its own. It's so pervasive that I find even addressing it to be refreshing, so thank you
Be angry more. I work in China but Im French, so people assume (and I nudge them to think), that it's a culture thing for me to express anger publicly at injustice or idiocy.
But it's liberating to be angry at bullshit (and God knows China is the bullshit Mecca), and AI is the top bullshit these days. We're not anti innovation because we say chatgpt is not gonna maintain our trading systems or whatever you work on. It's a funny silly statistical text generator that uses hundreds of thousands of video cards to output dead http links.
We're far from anything intelligent but it's at least indeed very artificial.
As someone who was in academia at the right time to really contextualize the seismic shift in the capabilities of automated natural language processing models that came of both attention and the feasible compute scale increase that allowed for them to use long enough context windows to outpace recurrent models in this regard, I really didn't think I'd end up having to roll my eyes at people falling over themselves to exaggerate the implications, but in retrospect it's clear that this was more me being unconscionably naive then than it being that unpredictable
Cause getting angry at a problem and spending 2 days coding to completely replace a long standing issue isnt something that happens...
People need to be less precious. You cant be happy all the time. In fact you probably should not be (life without contrast is boring). A little anger about the indignities and bullshit in the world is a good thing. As long as you're still rational and receptive, it can be an effective tool for communicating a point.
Or just communicating the emotion! I think aligning on an emotional layer of perception is important for shaking people out of automated behaviors when it's necessary to, and I dislike this cultural shift toward punishing doing any of that from the standpoint of its mechanism design implications almost as much as I hate it on vibes
Not only that, but the thing is that it’s all fake in our industry and the companies that we work on. People seem to be very sensitive today to showing any kind of actual emotion or feelings, be it anger or frustration. Everyone puts on the fake american service industry smile, say words like I hear you, we’re a team, we must be constructive. Then in the background all do the most insane political backstabbing, shit talk about other teams, projects, people walk over the careers and the future of others just to advance themselves, but as long as you put a smile on your face in the meetings and in public, none of that matters.
I mean you make some very good points but you sound like you could be kind of mildly upset if I squint at it right so I think you should really be more mindful and adopt a positive additude before I will even consider listening to anything you have to say
If you want a theory; a man who isn't in control of his emotions can present anything up to an immediate mortal danger to the people around him (particularly if they are female).
Being able to control negative emotions isn't a nice-to-have trait or something that can be handled later. There is an urgent social pressure that men only get angry about things that justify it - a class of issues which includes arguably nothing in tech. Maybe a few topics, but not many.
Anger isn't a bad thing in itself (and can be an effective motivator in the short term). But people get very, very uncomfortable around angry people for this obvious reason.
> If you want a theory; a man who isn't in control of his emotions can present anything up to an immediate mortal danger to the people around him (particularly if they are female).
What emotions do you really control?
We expect men to suppress this emotion. And there's is 400k years of survival and reproductive success tied up with that emotion. We didn't get half a percent of the population with Ghegis Khans Y chromosome with a smile, balloons and a cake.
It's not like violence doest exist. But we seem to think that we can remove it just like the murder in the meat case. Are we supposed to put anger on a foam tray and wrap it in plastic and store it away like a steak because the reality of it upsets people?
It's to the point where words murder, suicide, rape and porn are "forbidden words"... were saying unlike, grape and corn. So as not to offend advertisers a peoples precious sensibilities. Failing to see this behavior is a major plot point in 1984.
I think we all need to get bad to the reality of the world being "gritty" and having to live in it.
1) If you are comparing people's behaviour to Genghis Khan, don't expect positive social reinforcement. The man was a calamity clothed in flesh, we could do without anything like him happening ever again.
2) Violence != anger [0]. I don't know much about him, but Ghengis Khan could have been an extremely calm person. It is hard to build an empire that large and win that many campaigns for someone prone to clouded thinking which is a point in favour of him being fairly calculating.
> What emotions do you really control?
3) In terms of what gets expressed? Nearly all of them. Especially in a written setting, there is more than enough time to take a deep breath and settle.
> We expect men to suppress this emotion.
4) As an aside, I advise against suppressing negative emotions if that means trying to hold them back or something. That tends to lead to explosions sooner or later. It is better to take a soft touch, let the emotion play out but disconnect it from your actions unless it leads to doing something productive. Reflect on it and think about it; that sort of thing.
[0] Although maybe I should not that angery violence is a lot more dangerous than thoughtful violence; angry violence tends to be harder to predict and lead to worse outcomes.
> If you want a theory; a man who isn't in control of his emotions can present anything up to an immediate mortal danger to the people around him
You cant posit this and then go on to try and claim Violence != anger.
> The man was a calamity clothed in flesh
Nice, well said!!! He was also likely brilliant. Its rare stupid people make it to the top!
I hope that Ghengis Kahn NEVER happen again...But I think society is just a thin veil between us and those monsters. The whole idea of pushing down anger is just moving us one more steep from that reality!
Okay but we're not talking anger that's expressed by violent behavior or even clear significant loss of control, I'm talking people on the internet can pick up the mildest hint of anger from your tone or even subject matter. As a woman and a pretty scrawny one at that, as well as being, well, obviously very opinionated and belligerent, I have experienced every flavor of the threatening behavior you're invoking and I can assure you this has nothing to do with why people reflexively dismiss people who they think are being "emotional". More and more, the accusation of being angry specifically seems to be all people think they need to say to smugly claim to be speaking from a "rational" high ground, often despite having contributed nothing of substance to the topic at hand. Like pointing out that this person's blog post aimed at no one particular person did not really have to contend with the perception that this person was going to actually become violent at anyone, although actually I could see getting that impression from this post more than most, since it frequently explained the anger as cartoonish threats of hypothetical violence. I'm not exaggerating. When I see this in person and can make better assumptions about the genders of the people involved, this seems disproportionately likely to be leveraged against women, as are most arguments to "obvious" or "apparent" disqualifying irrationality, and this is not a shock because we are within living memory of much of work culture treating it as conventional wisdom that this should be assumed of all women by default. People really be trying to win epistemic pissing contests by posting something that looks like running "u mad" through google translate and back once, unironically, just as surely as you're trying to do that obnoxious thing of trying to invoke the gravity of situations in which people genuinely fear for their safety, hoping that gravity will somehow make it harder to question what you said for fear of seeming chauvanistically oblivious or whatever that's supposed to do
I propose the alternate theory that as in-person interaction becomes a smaller portion of most people's social experience, many have gotten worse at handling even mild interpersonal conflict without the kind of impersonal mediating forces that are omnipresent online, and this kneejerk aversion reaction can rationalize itself with the aid of this whole weird gilded age revivalist-ass cartoon notion of "rationality" that's become popular among a certain flavor of influential person of late and, especially in a certain kind of conversation with a certain kind of smug obnoxious person, seems kind of like classic Orwellian doublespeak
Also this position that "arguably almost nothing" in tech warrants anger seems super tonedeaf in a context where most of the world has become a panopticon in the name of targeting ads, you need a mobile phone owned by a duopoly to authenticate yourself to your bank, and large swaths of previously functional infrastructure is being privatized and stripmined to function as poorly as the companies that own them can get away with while the ancillary benefit of providing employees with subsistence and purpose wherever possible, while still managing to nickel and dime you for the privilege with all manner of junk fees, and offer poorly-designed phone trees in place of any meaningful documentation or customer service
Just going through your last paragraph; the logical implication of getting angry about any of that is either living in a state of ignorance or getting angry all the time. Either of those options is far inferior to just taking note of what is happening and calmly suggesting some improvements or working to make things better when the opportunity arises.
And these issues are just minor compared to all the terrible stuff that happens routinely. If we're ranking issues from most to least important things like "you need a mobile phone owned by a duopoly to authenticate yourself to your bank" are just so far down it is laughable (the wry type, like "why do I even care"). The fact that you need a bank at all is a far more crippling issue. Let alone all the war, death, cruelty and disinterest in suffering that is just another day in a big world.
Two things can be true at once. We live in a big world and in that world, there are many things that warrant our anger, some of which are more important or urgent than others. Yes, it's probably more important that there are two wars going on or that the rich country that I live in has become a police state that jails millions of people on dubious and often bigoted pretenses or that the capital that owns the industrial capacity that won the last major era of technological progress is hell-bent on continuing business as usual in a way that we're now pretty sure will drastically harm the ecological infrastructure we depend on to survive, and has been engaged in decades of attacking the scientific and political capacity to dismantle them. Also, many of these problems are directly aided and abetted by the owners of the current wave of technological advances, who have also created and continue to iteratively worsen a pervasive network of surveillance and control, as well as an experiential environment that reliably produces apathy and learned helplessness, while destroying significant hard-won freedoms and infrastructure in the process (including uber rolling back labor rights gains, amazon crippling public delivery infrastructure it views as competition, etc)
Epictetus wrote of concerning oneself more with that which one may be able to control than that which one can't, and people who aren't familiar with the Enchiridion have nonetheless internalized this wisdom. It pops up in lots of places, like in various schools of therapy, or in the serenity prayer. My career is in computers, and this website is a nexus wherein people who do computers for a living gather to discuss articles. Therefore, the shared context we have is disproportionately about issues surrounding computers. We are all of us likely better positioned to enact or at least advocate for change in how computer things are done in the world, and in each of the last 7 decades this has become a larger share of the problems affecting the world, and anger is difficult to mask when talking about problems precisely because one of the major ways we detect anger in these text conversations devoid of body language or vocal tone is expressing a belief that something is unacceptable and needs to be changed
I know someone that got bullied for taking leave to care for their special needs kids, and the positivity people came pouring out of the woodwork to accuse her of being confrontational when she was understandably upset. Not that the commenters above have indicated that they'd do anything like that, but yeah, it's WILD out there.
Author here. I spent Thursday evening with my vocal teacher, where we mostly giggled because I couldn't hit the F required to start Billy Joel's "Movin' Out". So yes, it was most decidedly humorous hyperbole, and it has been taken this way by like, everyone that's not on Hackernews. Genuinely astonished that it isn't as obvious to other people as it was to you, but something I learned quickly is that you simply can't control how your words are perceived, and it isn't possible to waterproof sufficient long-form content.
To be honest, this is kind of nice, because my girlfriend recently told me that I'm "extremely unthreatening" because of all the improv theater, fencing, music, reading, and writing. Now I know at least a few people on the internet are threatened by me. I'm a loose cannon, on the verge of totally unrestrained violence. I'm two steps removed from a dinosaur, and if someone looks at me funny, who KNOWS what'll happen.
I remember ~6 years ago wondering if I was going to be able to remain relevant as a software engineer if I didn't learn about neural networks and get good with TensorFlow. Everyone seemed to be trying to learn this skill at the same time and every app was jamming in some ML-powered feature. I'm glad I skipped that hype train, turns out only a minority of programmers really need to do that stuff and the rest of us can keep on doing what we were doing before. In the same way, I think LLMs are massively powerful but also not something we all need to jump into so breathlessly.