The AI wildfire is coming. it's going to be painful and healthy
ceodinner.substack.com
ceodinner.substack.com
I wonder for how long this will keep working. Can't be too hard to prompt an AI to avoid "tells" like this one...
I've been prompting the bot to avoid its tics for as long as I've been using it for anything; 3 years or so, now, I'd guess.
It's just a matter of reading and understanding the output, noticing patterns that are repetitious or annoying, and instructing the bot as such: "No. Fucking stop that."
People who put the effort into checking the output aren't necessarily checking more than style, but some of them will, so it will still help.
If you're getting a lot of value out of LLM writing right now, your quality was already garbage and you're just using it to increase volume, or you have let your quality crater.
Biggest one in this case, in my opinion: it's an extremely long article with awkward section headers every few paragraphs. I find that any use of "The ___ Problem" or "The ___ Lesson" for a section header is especially glaring. Or more generally, many superfluous section headers of the form "The [oddly-constructed noun phrase]". I mean, googling "The Fire-Retardant Giants" literally only returns this specific article.
Or another one here: the historic stock price data is slightly wrong. For whatever reason, LLMs seem to make mistakes with that often, perhaps due to operating on downsampled data. The initial red-flag here is the first table claims Apple's split-adjusted peak close in 2000 was exactly $1.00.
There are plenty of issues with the accuracy of the written content as well, but it's not worth getting into.
It's more things like AI delivering pizza that's under threat. You know, the actual value.
No, no, no! Stop that! The em dash is an wonderful little punctuation mark that's damned useful when used with purpose. You can't turn it into some scarlet glyph just because normal people finally noticed they exist. LLMs use them because we used them, damn it.
For god's sake, are we supposed to go back to the dark ages of the double hyphen like typographic barbarians in the hopes that a future update won't ruin that, too? After all the work to get text editors to automatically substitute them in the first place?
What's funny is that, when people first started noticing that LLMs tended to like the em dash, I'd mentioned to a friend that I hoped—rather naively—it might lead to a resurgence and people would think to themselves "huh, that looks pretty useful." Needless to say, I got that one wrong. Are we really going to sacrifice the poor em dash just because people can't come up with a better signifier for LLM text?
Em dashes are strong; they can take the abuse. While I'm right there with you on dreaming of a renaissance in quality writing, that's likely more fantastical than my own hopes that we'd see a resurgence in quality em dash usage. At this point, I'd probably settle for outright stagnation in writing ability.
Edit: I forgot HN strips emojis.
It seems like right now the most rational move to stay in the industry is to milk the AI wave as much as possible, learn all of the tools, get a big brand name on one's resume, and then land somewhere still-alive once the AI music stops? But ultimately if nothing outside of AI is growing, it's one big game of musical chairs and even that might not save you?
SV & big tech engineer money.
Majority of engineering fields do not make that kind of money to retire at 50. Comfortable compared to the rest of the country, sure.
That said if you’re making $250k+ a year and not on track to retire by 50, seriously please open a retirement calculator and figure out what you need to do to get there.
That said, a lot of people in US tech can probably retire relatively early if they run the numbers and don't have a lot of external expenses.
A semi decent apartment in SV will cost you ~$3k
Bills(phone, internet, electricity, etc) another $1k.
If you are married, groceries at least $1k.
Even if we assume you don’t do anything else in life, and you are in perfect health best case scenario would be $6k savings a month or $72k a year.
It would take you 10 years to save $720k plus whatever you make from investments.
That’s not enough to even buy you a house in SV. How are you going to retire?
Unless you assume you will get $250k straight out of college and keep up salary raises for 25 years.
Sure, if you don’t have kids, age with no health problems, never enjoy anything in life, you may be able to retire at 50 in Thailand or Philippines.
I think people commonly underestimate how accessible this stuff is
It’s easy to make a 40 year forecast spreadsheet for retirement, including housing costs, property, taxes, maintenance. Include vacation, budget, food, general cost of living.
I mentioned Lifestyle creep before but what is with everyone’s fried brains?
A small condo in a nice neighborhood in Santa Clara is below $500k. Yes, that’s a lot, and you certainly can get more bang for your buck if you’re willing to do a little commuting.
Btw a $1m house is accessible if you make $250k yr, although to be honest, I would highly recommend against it
I said "because you need to have a downpayment".
You reply "downpayments aren't that high".
Unless you're getting loans zero down, you literally still need to save to have your downpayment. While you're renting.
So where is my brain fried?
Even on a $500k condo, you're putting 10% down, you still need to have that saved up. Noticeably more, in fact, because I'm sure you'd agree "lemme sink every saved cent I have into my house downpayment" wouldn't be wise.
Many places.
First, you can get a zero down loan at that income level. Wouldn’t recommend it but I can tell you haven’t seriously explored home loans if you don’t know this.
Second, if you make $250k a year and you’re struggling to put $10k, or hell, even $50k down… you need to make a budget because damn
> you literally still need to save to have your downpayment. While you're renting.
Just so you know, I did it. I bought a condo a 20 minute train ride from sf, making quite a bit less than 250k.
I did it at $180k a year by moving into a studio in oakland, and that was overkill tbh. Saved $1.5k a month and biennial vacations were to New York and Vegas instead of France and Japan. Such a sacrifice /s. Now imagine how easy it’d be with an extra $70k (about the average annual salary).
The best advice i can make to you, if you really are making above 200k a year, is to make a budget spreadsheet.
I knew a lot of people who did almost exactly that ~18 years ago. It didn't go well for them.
And then it turned out that staying flexible as a renter and setting aside cash set me up to buy after a correction instead of before. That part went very well for me.
Be careful with the assumption coded into your "forecast spreadsheet"
There is no one size fits all solution but i’m surprised at how many people here are inadvertently revealing to me that they haven’t even tried evaluating.
For example, you saying there’s nothing “affordable” when the baseline assumption is an income of $250k? Can tell you haven’t looked at what’s in your price range. Alright, good luck I guess!
I don't see how we get to "why would you rent and save for a house at the same time?" from "it takes 10 years to save $720K" but whatever.
Assuming a 4% draw down (conventionally agreed to be safe) is over 5.5k a month.
Also $1k month on bills? Groceries too?
Judging by your inflated costs for everything, and ann idea that a house (versus more modest accommodations) is what the goal is, you’ve got Lifestyle creep. And, things certainly get a lot easier when your spouse also works.
Edit: all % numbers are per year
Consider the case of condos in cities. If you were to buy outright, you effectively get a return by not paying rent (i.e. paying yourself rent). Rent is usually ~5% of the condo cost. HOA + property taxes is 2-3% so subtract that from the rent return i.e. net return 2-3% (5-2/3%). The rest of the return is appreciation from the underlying real estate prices. I am excluding maintenance costs because they are negligible in condos.
On the other hand, if you rent and put the entire amount (that you would have paid to buy the condo), you get ~10% per year. To break even between the two scenarios, you would need real estate prices to grow 7-8% (2-3% + 8-7% = 10%).
Beyond this, there are psychological reasons to buy vs rent. Buying - ability to customize the space, peace of mind because of perceived stability etc. Renting - flexibility, peace of mind because of no long-term obligations etc.
A mortgage is an interpolation of the two cases at the cost of the interest one pays. It is noteworthy, at least in the US, that for most people, this is the only time they can borrow several hundreds of thousands at relatively low costs.
For a long while we managed to stay around $500-600 but that was before COVID inflation. I dunno how the official inflation rate's as low as it is, we don't buy much that'd be considered "luxury" level (we're not buying caviar, say, and rarely even get stuff like the grass-fed "fancy" butter [actually yellow instead of white, tastes like something rather than just having texture but no flavor] instead of the cheapest available) and I'm pretty sure we buy a lot less meat per person than the US average, but if we fill up a cart now it's like $250-$300. I've hit $150 on small shopping trips where I didn't even fully fill one of the smaller, short carts.
$150 phone $200 electricity $200 gas $200 water $80 internet $80 trash
Car insurance? Gas? I’m ridiculously generous when saying you can save $6k per month.
I regularly frequent HN, and even comment from time to time, but I don't work in tech nor do I make bank. I'm a cashier at a gas station. Lol. I'm lucky if I make $16000 a year after taxes.
Generally outside SV:
- If you are making $250+ it is at least middle management (not tech work) AND
- Only in zones where cost of living is eating this up (e.g. UK/Europe/Australia/etc can get to this equilvalent salary but costs for example for rent, food, tax, etc are much higher).
In most countries SWE is above average pay, but it isn't life changing and it still unfortunately has the boom/bust cycles.
I've met some very good engineers who have built some great large scale solutions who are on less than this salary often in non tech firms being outside of the SV area due to personal reasons (e.g. can't move due to family, too old to do the interview dance SWE has become these days, etc).
Though, of course, if you're living from investment income you should be aware you're living off the work of other people.
Isn't the logical extension that everyone lives off other people?
This was basically the point of "you didn't build that" (https://en.wikipedia.org/wiki/You_didn%27t_build_that)
Unless you're working very obviously outside the blast radius of an AI-bubble correction (you'd know if you were) or are a very high-value VIP (again, you'd know), you should assume you'll be spending some time without a job within the next few years. Possibly a long time.
You might get lucky, but it's not really going to be in your control and "milking the AI wave, learning all the tools" isn't going to change your odds much. It really is musical chairs. Whether you lose your job will depend on where you happen to be standing when the music stops. And there are going to be so many other people looking for the same new chair as you, with resumes that look almost exactly like yours, that getting a new job will basically come down to a lottery draw.
If you think the AI stuff is cool, study it and play with it. Otherwise, just save money and start working on the outline for that novel you've been thinking about writing.
My plan as someone who was thinking of leaving tech anyway (remote work is not for me, and practically any new tech job I get will be at least as remote as this one has become if not more so, and I want to program not manage programmers, artificial or otherwise) is to stay where I am pushing through to the other side if possible and if not, I'll find myself redundant. At that point I'll end up on a lower wage doing something else from the ground up, but if LLMs are going to be what we are told they are programming will become a minimum wage job for most anyway. Either way, sticking where I am for now, tightening the purse strings a bit, saving as much as I can, is the best course of action.
Tech and software's investment balance sheet comes down to a largely fixed cost of development vs. a large customer base where every customer has little to no additional cost.
If you manage to burn the bridges or at least scare hundreds of millions of those people into exploring alternatives, that really eats into your total target market in the long run.
Is the return measurable and immediate?
Is it really?
Dentists offices that only need 1 receptionist instead of 2.
A dramatic reduction in front line tier 1 customer support reps.
Translation teams laid off.
Documentation teams dramatically reduced.
Data entry teams replaced by vision models.
Large banks have tens of thousands of call center employees and a large % of calls they handle are perfectly solvable with a good AI bot. They are working very hard to cut call center staff as quickly as possible.
People don't realize how much a call to customers service costs. Back when I was at MSFT, a call to tech support for our product costs $20 to have someone pick up the phone. Since we were selling low margin HW, a single call to tech support completely erased the profit from that product's sale.
Layoffs have already happened and they will continue to happen.
One can argue this is a positive, as a customer if I can push a few buttons and issue a voice command to an AI to fix my problem instead of waiting on hold, that is a net positive. Also the price of goods will drop since the expected cost of customer service factored into the product price will drop.
E.g. $30 / support call, 1 in 10 customers call support during the lifetime of a product, $3 saved, but the way costs are structured, $3 saved in manufacturing can end up as nearly $10 off the final retail price of a product.
(And in competitive markets prices do drop when cost savings are found!)
>One can argue this is a positive, as a customer if I can push a few buttons and issue a voice command to an AI to fix my problem instead of waiting on hold, that is a net positive.
If you could do it through the website then you would be much happier than having to argue with a chatbot. And if you can't do it through the website, there aren't going to let a robot do it on your behalf.
Out of the things you listed the only ones that seem plausible are translation team and data entry team, though even there, I'd want humans to deslop the output.
Just a couple days ago a scheduled a furnace repair through an AI receptionist on the phone.
Layoffs in tech support and customer service already happened last year.
Entry level sales jobs doing cold calling have been replaced all over the place.
> The return is measurable and often immediate, not hypothetical.
It's one thing to let go some people and replace them with AI.
It's quite another to have a measurable and often immediate, not hypothetical return on that decision.
I have used AI translation professionally for a few years, and between hundreds of people in long conversations, nobody has ever asked if the text has been translated. Before AI translators, you could write at most one message and people would notice.
The chatbot, acting as my agent, whether on the website on on a call doesn't have more permissions than I have.
They're saving pennies but at what cost?
But not in the sense of singularity and explosive intelligence, but in the sense of a flaming explosive bubble of slop
Today they are a durable asset functionally, longer than they are economically. So there is no reason in a market with less demand, that their economic payback windows cannot be extended further into their functional lifetimes.
There will be energy cost incentives to replace GPUs. But turnover can respond sensibly to demand as it revives, while older GPUs continue working.
Also, the data centers themselves, and especially any associated increase in power generation, will carry forward as long term functional value.
I doubt any downturn in compute demand lasts long. The underlying trend, aside from AI, was for steady increases in demand. Regardless of bad AI business models, or investment overhangs, a greater focus by more entities on AI product-market fits, along with cheaper compute, will quickly soak up cycles in new and better ways.
The wildflowers will grow fast.
their huge customers will be able to produce ASICs hat will be faster and cheaper to operate than their GPUs
jensen has to be the luckiest man in the world, first crypto, now "AI"
BTW, stock price is not everything, Cisco survived, grew, and it's the backbone of internet today.
Are we sure this will be the case? Perhaps the sweet spot for hardware that can train/run language models is the GPU already, especially with the years of head start Nvidia has?
Why? NVIDIA is better positioned to produce faster and more efficient ML ASICs any of their huge customers (except possibly Google). And on top of that, the fact that there is a huge library of CUDA code that will run out of the box on NVIDIA hardware is a big advantage.
Arguably, this shift has already happened. Modern NVIDIA datacenter GPUs, like the H100, only bear a passing resemblance to a GPU -- most of the silicon is dedicated to accelerating ML workloads.
As a possibility for example I can see them transforming from a GPU based corp into a parent company for many full or partially owned "subsidiaries". They still manufacture chips to be "vertically integrated" but that becomes bread and butter as an enablement rather than the main story (e.g. Google TPU's). As their margins go down the value accrues to what they are owning (the business units/product areas).
Meanwhile, 67% of my time is gone fixing autocorrect on apple devices.
You: No one knows anything AI helps with.
Yeah, okay, if you ignore everything every user says then it is indeed a mystery.
The only thing I trust about these right now is my own experience
https://xcancel.com/elonmusk/status/1997307084853870793#m
One can only imagine the amount of covert promotion.
I don't have much proof, but given the incentives and the possibility of doing it I'd be surprised if it wasn't happening everywhere. How much would be enough to pay the top 50 influencers in a market to push something? To hire 100 people to be active full time on all social media sites? To a company with billions they wouldn't even notice the expense
Default to skepticism and double down on your critical thinking skills. More important than ever today
You might not like using LLMs. You might not find them useful. You might think they're bad and harmful (I do). But to claim that no one finds them useful is a completely different position, and one that's about as disconnected as it's possible to be.
I have all of those. Most don't use AI at all. Some use it on a limited basis but it is unclear if there is any worthwhile gain in productivity. Remaining are two who use it with regularity, including one who's all in. I personally use it for 2 limited use cases. Sometimes it helps. Sometimes I'd be done sooner without it.
Conversely, I need to mediate an epidemic of AI foistware and AI UX pollution. 100% of my userbase is subject to overpushy AI offerings and an endless minefield of shifty, unwanted AI elements. These users are clearly more productive when I keep AI out of their way.
On balance, AI is presently a net negative for my clients.
The only real difference this time around is all of the datacenters being built. There's real hard asset costs making it much riskier and capital intensive.
Sure, there might be hundreds or thousands of small startups in the AI game, and some are probably as viable as the fabled Pets.com. But even if they all crash and burn, it's going to be a rounding error compared to the 7 companies I mentioned above. The AI will be alive and kicking, and nobody will even notice.
The companies that are sustainable with their own revenue, covering their runway or nearly there, are likely to be alive if there’s not investors to keep them alive. Those with ridiculous commitments expecting a hail mary until their business model materializes, are living on borrowed time
I think the best metaphor will be the California gold rush. There is definitely gold there but most of it has already been mined. The people who are entering at this point are woefully unprepared, assuming that they can vibe their way into a fortune, when the rest of the gold requires hard earned labor.
Gold, tulips, real estate, rail, witch hunts, satanic panic..
We shouldn't be worried because "this time it is different".
"AI" is none of those things. It is "totally different this time". "AI" is going to do all the work for us, we are all going to get UBI then at some point as it grows the "ASI" is going to either figure out how to grant us immortality or cause mass human genocide.
It is all completely rational this time.
Har not to see parallels with the current AI bubble.
I don't see those companies promising that. They are bragging about being able to replace jobs for millions people and their CEOs are also simultaneously taking pretty dismissive attitude toward lesser people. If is more of "we will replace millions of people of eff you, billionaire class will become more powerful.
I am not saying it will realistically happen, I am saying that is what current messaging is.
But that was the same thought for me. The other totally missed aspect: a fire kills all life in area.
This is flat out wrong.
I'm sure some of them have managed to shake some change out of the VCs but these wanna-be shovel sellers are just gonna let their domains expire and move on to the next scheme with little overall damage to the economy.
I have no idea about their financials. They just annoy me, because they mask their ads as posts/comments. And use ChatGPT to generate those, they are like 2 page long drivel.
Interestingly this time around I could see the 'fire' affecting mid-large corporations (or at least some divisions of them) if they don't adapt. Adobe, being heavily focused on graphic design seems like it could be under pressure. Low-end consulting / outsourcing is largely doing the same work AI is good at. Similarly with technical gig-work (like Upwork).
They have no business case - they are 'burn money and hope AI allows us to build something we can monetise'. That's not a business model.
I think a huge breakthrough for AI was priced in, and we are still waiting to find out if it will come and what it'll be.
Personally, as this article seems investment focused, I see no downside to diversifying away into more varied kind of investments, but then again, I'm not a pro, so take it with a grain of salt..
This list of companies made me wonder a bit. Technical progress has been huge, no question about that. But as for the actual quality or experience for the user/customer - I have the impression everything got worse, starting from Google from the first wave.
Then the cycle is broken, and there might be no survivors, or the regrowth may be so far into the future that it will make no difference for most of the survivors.
This is only because businesses are full of folks with short-sighted FOMO desperately trying to cram AI features into any product they can. AI is the new digital clock.
We've put so many layers between the engineers and customers and diluted any accountability to demonstrate positive ROI—even if it's theoretical—that we do pointless work for nobody. I'm not going to complain too much personally because all those layers make it possible for me to just pull cards and collect a paycheck but I'm surprised nobody on the business side even somewhat cares if the work they're paying for is worthwhile.
Who has gained from the efficiency? We haven't gotten more vacation days and we haven't gotten more share of the money.
I think it should be natural that jobs end up being mostly pointless. Why should we produce exponentially more value without getting a share of that value?
But your money buys stuff that 50 years ago would have been too expensive for the richest men in the world. A pocket supercomputer, advanced diagnostics and medicine, instant access to information anywhere in the world.
the number of expenses required to minimally live
(ex:utilities, transpo, insurance, comms) and
the ever escalating costs of those added requirements
More to the point, a pocket supercomputer is an irrelevancy for a typical wage worker, who's earnings are far insufficient for even the barest self-sufficiency.
Frankly I've added some of the features of my own initiative. They were low hanging fruits and really helpful in some cases, and in others they are placeholders waiting to be better integrated or expanded depending on the users requests. Nobody forces anyone to use them or even notice them, so why not?
As I said: these features look like magic in demos, it's not because of the hype that managers want them integrated but because of genuine enthusiasm. But they require more development and maintenance effort than was apparent from the demo. Also, there's a clear discoverability problem due to the fact that an agent has basically no UI.
If anything, you should accuse them of foolhardy recklessness. They are not the sticks in the mud.
That's a nice crystal ball you have there. From where I'm standing, model performance improvements have been slowing down for a while now, and without some sort of fundamental breakthrough, I don't see where the business value is going to come from
That's a fairly tall order.
Companies that are hastily shoehorning AI into their customer support systems could instead devote resources to improving the core product to reduce the need for support.
I think that is the recklessness in question. Throw in that there is no profit for OpenAI & co and that everything is fueled by debt and the picture is grim (IMHO)
The key is for them to build before the money runs out--I'm not sure they will have enough time.
Musk is planning for 1 megaton/year of satellites, each with 100kW, yielding about 100GW per year.[1]
He thinks they can do that in 4 years, but adjusting for Elon-time, it's probably no less than 8 years, if ever.
But will the AI money last that long? Maybe not.
-------------
So about 2.5 MW of solar potential ? That's indeed quite impressive, but for serious compute a lot of energy needs cooling will eat into that.
Also building all the radiators and structure from in-space resources could be quite substantial energy investment, same with the energy to put it in the final orbit.
See also: the big deepseek smear campaign.
Man that is the question, isn't it
The GFC doesn't have anything in common with the .com bubble, maybe we'd have see another tech bubble in the first 2010s if there were not a GFC, but it's fundamentally wrong to place those 2 things nearby.
It is not that simple. You need to consider factors outside of the silicon valley, outside of USA, outside of technology business. There is lot of world out there. These analogies and predictions don't come out into the global scene to have a look at the what's going on across the globe.
The bubbles which happened earlier are not insulated phenomena that happened in silicon valley labs. It is a complex interaction between various forces.
For example, social and political norms may turn against all that is AI. Any AI-enabled service or product might be seen as serving plastic food.
This is some of the most nihilistic thinking I've read linked from this site. Also it lacks the nuance that when brush is left to accumulate over due years of forest mismanagement, the resulting wild-fires are much worse and unnecessarily destructive in ways that are hardly describable as "ecosystem health."
I note your AI missed the crypto hypecycle. Maybe because it really was a bubble.
It is not a bubble. It is not a fire ( cleansing or otherwise ). It is, however, a piece of technology that is, misguidedly, plopped hard into everything without regard for what it is actually good at. This is why I despair when I see AI in notepad or "ai protects okta'.
I am concerned, because I do see a big change on the horizon coming, but it is not the change that is being presented. It may not be the feared ai/agi/asi ( depending on one's particular bent ),but rather deep re-entrenchment of existing ecosystems in ways that will make things a lot more difficult overall.
Here is what I mean by this:
- the internet as we once knew it, is effectively dead - the ones who can ( money-wise and knowledge-wise ) and see the need to, move behind local networks - those that can't ( money-wise, knowledge-wise, or circle-wise ), are forced into locked systems that effectively become AML for... anything ( and if you did not experience it yet, I am assuming you did yet try to buy anything that has -- lets call it -- dual use )
It is bifurcation ( or what some media call k-shaped these days ), but it is not a fire at all. If anything, these are very, very aggressive vines.
Maybe it's simply less visible?
I have no account at any of the social media giants (except HN but I think that does not count). I mostly use the Fediverse and specialized forums. I would argue that it feels similar to the "old" internet.
"AI" hardly has any working products. Vibe coding is foisted upon companies by CEOs who want to promote their friends' products or who want to use it as an excuse for firing people or who have circular revenue agreements with other companies.
This is like the housing bubble of 2008 which was based on hot air and incorrect algorithms.
They can write it off and move on to other things. But that is not what the new wildfire talking point says. The wildfire framing says that the underlying tech is as valuable as the tech of 2000 was.
I wasn't a fan of any Microsoft products at the time, though Excel was pretty good when I started using it heavily a few years later.