[cmiles8]

This is important context in the wake of yesterday’s “raise” announcement. A lot of this stuff seems to just quietly never happen once the ink on the PR puff dries.

The AI industry increasingly looks in scramble mode to keep the hype going as those storm clouds of financial and business reality get darker and darker on the horizon.

[cj]

For a company bringing a new technology from zero to mainstream, I think it's pretty normal that there will be a lot of failed attempts at productization.

The thing that isn't normal is the degree of experimentation relative to company valuation. Normally once a company reaches $700 B+ valuation, they've figured out their product and monetization strategy. ChatGPT is clearly still iterating heavily on that - not normal for a company that size.

[mandevil]

And not normal for a company that has been at it this long.

The Apple II went on sale on June 10th, 1977. Visicalc went on sale October 17th, 1979- 860 days separate the two. ChatGPT was opened to the public on November 30th, 2022, which was 1219 days ago- almost 50% more time has elapsed than between the Apple II and Visicalc.

[ianbutler]

Without me trying to be snarky why do you feel spreadsheet software launching is comparable to this scenario?

[mandevil]

Visicalc is often described as the killer app of the first generation Personal Computer(1). It was the product that drove them into every small business in the country, that blew up sales of personal computers and brought them out of the realm of hobbyists into enterprise. And, honestly, I think Visicalc and spreadsheets are still a greater benefit than what I've seen out of generative AI today. And that happened a lot faster than where we are today with generative AI. Apple had enormous actual profits by 1980 (Apple IPO'd in 1980 with a 21% operating margin). So I think that a lot of the "just got to give it more time" argument misses that the previous computer based revolutions that we know about productized and threw off gobs of cash a heck of a lot faster than this one has.

If the end result of this is "certain classes of white collar workers are 10-25% more productive" (which is the best results I can extrapolate from what I've seen so far) then it's really hard to imagine how OpenAI can return a profit to their investors.

1: https://en.wikipedia.org/wiki/VisiCalc#Killer_app is pretty much the normal narrative on Visicalc and its importance to the Personal Computer.

[ApolloFortyNine]

>If the end result of this is "certain classes of white collar workers are 10-25% more productive" (which is the best results I can extrapolate from what I've seen so far) then it's really hard to imagine how OpenAI can return a profit to their investors.

If we take this as face value, and say that the absolute best case scenario is there are literally no other uses for AI but helping programmers program faster, given 4.4 million software devs, with an average cost to the company of $200,000 (working off the US here, including benefits/levels/whatever should be close), those 4.4 million devs with 20% productivity would save roughly 176 billion dollars a year.

Some companies will cut jobs, some will expand features, but that's the gist. And it's hard not to see the magnitude of improvement that's come in just 3 years, though if that leads to a 'moat' is yet to be seen.

[notahacker]

> If we take this as face value, and say that the absolute best case scenario is there are literally no other uses for AI but helping programmers program faster, given 4.4 million software devs, with an average cost to the company of $200,000 (working off the US here, including benefits/levels/whatever should be close), those 4.4 million devs with 20% productivity would save roughly 176 billion dollars a year.

I don't think that's necessarily out of line with struggling to return a profit to investors though: an individual company is only ever going to capture a tiny fraction of the productivity improvements it enables its customer base to make[1], its own cost base is unusually high for tech, and investors are seeking a 10x+ return on an $852B valuation for a company that isn't even the market leader in that segment (which isn't the only segment, but it's the optimum B2B one). You can have a great business with a great value proposition and a sustainable moat and still not generate the desired returns on investment at a $852B valuation.

[1]and that's productivity improvements over the best-known free models, not productivity improvements over reading StackOverflow

[ianbutler]

Thanks for the in depth explanation. I was definitely not up on my tech history here. :)

[mandevil]

Sorry, I forgot that for many engineers this is, in fact, their first time going through a technology cycle like this, and so would need more explanation. I am too young for Visicalc myself, but the cycle that I saw while I was in high school- the dot-com bubble- doesn't have convenient, easy to mark out dates like the PC does.

Thinking... Thinking... Tim Berners-Lee proposing HTTP in 1989 is kinda like the original Attention is All You Need paper, I guess? Netscape 1.0 release in December 1994 is ChatGPT 1.0? And then Amazon.com opened up to the public in July 1995 and then IPO'd in May 1997 (after raising less than 10 million dollars in two funding rounds). But once again we have the business side of these previous cycles moving much faster than this one.

[bombcar]

I took it the other way, spreadsheets shook up the world way more than AI has (to date) - it's possible that history will look back and count AI as the bigger "thing" but if I had to pick a killer app, VisiCalc and computer spreadsheets in general would beat ChatGPT.

[aworks]

Good link from the other day about spreadsheet impact on business https://davidoks.blog/p/how-the-spreadsheet-reshaped-america

[taco_emoji]

Visicalc is widely regarded to be the first "killer app" for the Apple computer. Perhaps even the first "killer app" period.

[Marazan]

VisiCalc was the killer app.

[zozbot234]

VisiCalc was "the" killer app for early micros, but being able to edit a written text on screen and then print it out with letter-like quality was nothing to sneeze at, either. This was plausibly a key gain in efficiency for the service sector, perhaps comparable to the 10%~25% that's now being talked about re: LLM's (which is huge on a secular basis).

[ianbutler]

Ah got it. I wasn’t drawing that connection. Thanks

[skydhash]

IMO, the AI companies are trying to be both T-Mobile and Google Doc at the same time. Even Apple is struggling with being both the platform and the product. The issue with OpenAI is that the platform has no moat (other than money) and the product can be easily copied. In the game console world, the platforms have patents and trademarks, and games are not easily produced.

[aworks]

I'm a retired engineering manager so judge me appropriately :-) I've had 1000 good chats with ChatGPT on a wide range of topics. I build personal Excel and Access applications but not any real programming. I don't need workflow automation although I will dabble with Codex. I'm curious why I should abandon what works for Claude.

[skydhash]

You shouldn't. No need to rush to buy a TI-84 to do simple arithmetic. I don't use either because I can learn just find from docs and textbooks. And I don't have that many problems to solve with computing.

[beachy]

The Apple II was so simple (by today's standards) that it came with a complete printed circuit diagram. Visicalc was so simple it was written by two guys in a year.

AI is so many orders of magnitude more complex that the comparison is not really useful.

[mandevil]

This complexity requires a lot of money- from investors- to sustain. If those investors don't see a return on their investment before they get too anxious, then no more money will be invested and the business is dead. So that would suggest that there will be even less patience from the money than the investors in Apple had. If you are correct that this greater complexity actually makes it harder to productize, then it is hard to see how frontier model generative AI will be viable under a VC funded domain.

It is entirely plausible to me that there are great technologies that are impossible to reach via the normal means of VC/investor financed capitalism. I certainly have encountered market failures requiring extremely patient money (usually in the form of government subsidies) to produce a useful product that eventually does have market value. That has worked many times in the past. But so far generative AI has not had that, and looking at my non-technology friends, I very much doubt that there would be much support among them for government subsidies of AI companies. AI companies have made too many people unhappy, served as too much of a punching bag, to be in a good position politically for that.

[convexly]

For most companies an $850B valuation means you figured out the business model. Here it seems to mean you convinced enough people that figuring it out is inevitable. Those are very different bets.

[aworks]

I'm curious how long it took Google to reach that valuation.

[ai-x]

Which is a good thing. Elon has showed the world, the only thing that limits the upper bound is bureaucracy, extreme risk-averse and no culture for experimenting.

More and more companies will start operating on the correct reward/risk curve or else getting crushed by firms who do. OpenAI has forced Google, Apple, Meta out of their comfort zone because they know OpenAI will eat their lunch

[jmye]

Literally every part of this comment is confusing. Elon hasn't shown anyone anything interesting in at least a decade. OpenAI hasn't forced Apple to do anything - LLMs aren't impinging on hardware or bundled services, and this literally seems right up Google's alley (and they're arguably better at it than OpenAI has demonstrated, now that first-mover-ish is long past).

I suppose Meta's recent comfort zone was simply a stupid bet on VR, so sure, maybe one part of the comment isn't confusing.

I don't understand what you think you're seeing.

[mrhottakes]

True, Elon has really been achieving win after win with Tesla and Twitter.

[dgellow]

Anthropic does look healthier, with their enterprise focus. Or am I missing something?

[cmiles8]

They nominally come across as a more stable ship with less clouds over its leadership.

However all of the major privately held AI players are struggling to paint a business and financial picture that doesn’t look “terrible” at best and “verge of market moving implosion” at worst.

For now the only thing keeping this all alive is more and more irrational cash being thrown on the pile in the faint hope that something stops the implosion from happening.

[zer00eyz]

Outwardly it looks much better.

But between their token curtailment and time of day restrictions, and some of the clues in the code leak (regex for sentiment, telling the public client to be "brief") it seems like they are facing some capacity issues.

Im guessing that the accountants at all the AI incumbents drink heavily.

[mpalmer]

Anthropic can't prop up Nvidia and the chip industry itself. If AI as an industry can't start turning a dollar into $1.05, a lot of stuff starts falling in value

[zozbot234]

Nvidia will be fine, same for the chip makers. They're selling shovels during the gold rush and they stand to gain even if local/on prem AI eventually dominates for casual users.

[darth_avocado]

> healthier

Correct. As compared to other AI companies. Tangible product, specific market segment and stable user base.

But whether it is worth a trillion dollars (like some of the peers are pretending to be) is yet to be seen. A lot of companies are using Anthropic products, but whether the spend is worth it, is also yet to be seen. A more realistic end state for Anthropic would be that they’d enterprise customers, with limited but steady spend due to Anthropic finally having to stop subsidizing tokens and a valuation in around $200-350B.

[bandrami]

Yeah TCO on AI is the big hazy uncertainty that a lot of large firms are still waiting on firmer numbers on before they really adopt. And in the nature of things we can't get TCO numbers for another year or so.

[adventured]

The LLM usage will generate hundreds of billions of dollars in ad revenue, which will be wildly lucrative in terms of margins (not as good as Google search used to be). If GPT is a leader in that, they'll take a sizable share of that pot.

There's a lot more money in being Google -> consumer ads, or Amazon -> consumer ads, or Meta -> consumer ads, than there is in being Anthropic -> enterprise.

Just take a look at the enterprise. Amazon's ad business alone is already a better business than Oracle or SAP or Salesforce, with superior margins, and it's growing faster too.

And of course everybody knows the Google & Meta ad monsters.

The only question remaining is who is going to extract all those LLM ad dollars, how will that break out. Right now it's Gemini and GPT in the obvious lead, with Anthropic in third, and Meta & Grok nowhere to be found (permanent situation for those).

[operatingthetan]

>The LLM usage will generate hundreds of billions of dollars in ad revenue, which will be wildly lucrative in terms of margins (not as good as Google search used to be).

This seems like ... not the situation we are in. LLMs are great for coding now but their text generation capabilities aren't exactly capturing the masses or replacing their jobs yet. People are already tired of the deluge of fake content on the internet, it's not going to drive a second revolution in web ads.

The $20-200 LLM plans are all subsidized and aren't paying for themselves. Something has to give here.

[steveBK123]

> The $20-200 LLM plans are all subsidized and aren't paying for themselves. Something has to give here.

Whats interesting to me as well as much as companies are pushing AI adoption, i have started to hear AI token spend limits enforced across a few companies, so its not entirely clear that b2b can make them profitable yet either.

If all the models reach good enough, then low cost provider would win. Gemini seems like a safer bet since Google controls more of the stack / has more efficiencies / cross selling / etc.

It’s not like “best” has won any other b2b arms race in the past.

[operatingthetan]

>If all the models reach good enough, then low cost provider would win. Gemini seems like a safer bet since Google controls more of the stack / has more efficiencies / cross selling / etc.

Gemini is the best deal too. For $20: you get multiple quotas per day across the products (web, CLI, antigravity, AI Studio) 2tb of cloud storage, and you can family share the plan.

[SpicyLemonZest]

I don't know Gemini's pricing model in detail, but in general pricing doesn't generalize well between personal/hobbyist and enterprise use. Consumer pricing of variable costs is a balancing act, and most Gemini users aren't going to be anywhere near the quota; a company of 1000 can't always buy for $20,000 what 1000 random users with $20 personal plans are theoretically capped at.

[cmiles8]

In large part because most companies have a set budget for IT spend. Thats how “normal” profitable companies operate outside this cash burning bonanza that’s going on.

And in that reality one can’t just magically spend a bunch more on some fancy new thing, especially when said fancy new thing isn’t retuning value. So “token limits” and cost controls on B2B is entirely expected here.

[steveBK123]

> especially when said fancy new thing isn’t retuning value

I think this is the key element. Either they can't measure the value, or it's far far lower than anyone wants to believe, or both.

I think the problem is less that it makes some coding tasks XX% faster, but that the end to end of a SWEs roles tasks is only improved by some much smaller Y%.

If a CTO sets $10k/year spend limits on $500k SWEs.. they must not believe any of the hype.

[adventured]

LLM usage will largely replace traditional search, and that's stage one. To be specific, search will be consumed by the LLMs, it'll be merely an aspect of what they do for the user, and that'll include handling the more intricate details of the search, refining the search, understanding the results of search, etc. The age of the typical user handling any of that is about to end. Search will more be a feature of Gemini in the not very distant future, rather than Gemini being bolted onto/into search.

Fuller integration into the user's life will bring ever more ad opportunities (and it doesn't matter if the HN base hates that notion, it's going to happen regardless). That'll happen over the next decade gradually.

Shopping, home management, tasks (taxes, accounting, lifestyle, reminders, homework, work work, 800 other things), travel (obvious), advice & general conversation (already there), search (being consumed now), gaming (next 3-5 years to start), full at-work integration (gradual spread across all industries, with more narrow expertise), digital world building (10-15+ years out for mass user adoption). And on the list goes. It's pretty much anything the user can or does touch in life.

[operatingthetan]

> To be specific, search will be consumed by the LLMs, it'll be merely an aspect of what they do for the user, and that'll include handling the more intricate details of the search, refining the search, understanding the results of search, etc. The age of the typical user handling any of that is about to end.

We already have the tech for that, why hasn't it happened? People are revolted by the AI results in Google. AI isn't going to make people use their computers more. It's not opening up a new consumer market. This is just making each search infinitely more expensive.

[eisa01]

I find searching chatgpt.com and asking for sources, then visiting them, works much better than Google to find niche topics

[satvikpendem]

Who is revolted? I use the AI Google results every day when asking for specific questions, I rarely visit the webpages before anymore. Also Google already injects ads into conversations in the form of Google Shopping affiliate links.

[satvikpendem]

You see it already with how many people use LLMs for everything these days. Google Gemini can also integrate with your other Google apps to personalize further, and Gemini already has product placement ads.

[cmiles8]

Google is already dumping LLMs into search and it works well and is free.

[sylos]

It doesn't work well. The searches are wrong and uninformative much of the time.

[mrhottakes]

It works very poorly

[zer00eyz]

> The $20-200 LLM plans are all subsidized and aren't paying for themselves. Something has to give here.

Expert systems were amazing. They were not cost effective.

There might be another bitter lesson to be had here, and unless the accountants start talking we're not gonna know any time soon.

[rchaud]

Google launched in 1998 and were running ads by 2000. Considering how much more access to adtech product talent there is for OAI a quarter of a century on, what explains their hesitation to pick that route and make billions? After all they had billions avaiable to acquire designer bauble maker Jony Ive's company.

[delecti]

The first AI company to cram their product full of ads will get roasted over the coals for it. My guess is they're all playing chicken and waiting to be the second to do it. I'd also guess that they're all already thinking about ways to introduce it that will generate the least backlash.

Google could do it in 2000 because their search was legitimately so much better, and also because their ads were comparatively more relevant and unobtrusive than modern ads. In comparison, LLMs are relatively similar in performance unless you're picky enough that you're probably already paying and thus wouldn't be in the ad-supported tier.

That said, I wonder if ads are even lucrative enough to move the needle relative to how much training costs are increasing with each generation.

[bombcar]

People forget it took Google years of frog-boiling to get us to where we are now.

The first AI to insert blatant ads will be dumped for some other model overnight. Look at the Copilot "backlash" over their "product announcements".

[df2sdf]

Yep. Finally someone who knows how to think straight.

Google built up immense surplus and ate into it slowly.

OAI can't suddenly start cramming Ads as it fights to survive. Im sure they are employing questionable tactics to keep users hooked. Won't really work on a large enough user base to make the economics viable given the competition they face.

[cmiles8]

“The LLM usage will generate hundreds of billions of dollars in ad revenue”

And yet every attempt to extract even minimal ad revenue has been canned to date as something nobody wants with AI providers retreating in failure.

I don’t doubt that there’s “some” ad revenue to be had but there’s little evidence that ads are going to save the day here.

[bdangubic]

These exact words were said tens of thousands of times about Facebook (am old enough to remember those discussions :) ), “no way they can monetize on mobile” (this was the most fun).

rules are simple, if you have Xbn or XXXm users on your system, you will make big bank in ads eventually

[iAMkenough]

At that time, Facebook provided a free service without any real competitors. The masses will switch to Meta AI or Gemini or Claude at the drop of an ad that annoys them enough.

[adventured]

Gemini, GPT and Claude will all have ads on the consumer side. They will go together in quasi lock-step into the ad future, because that money is gigantic and they're going to need it.

The masses will have no say in the matter. Just as they had no say in the matter with Google's ads getting ever more intrusive, or cable prices previously, or streaming prices going perpetually higher in the present, or YouTube ads, or anything else. Consumers will have no say in the matter, they'll take it and that's that.

With only three relevant competitors (maybe Mistral in Europe), there will be nowhere to flee the deployment of ads.

[harmonic18374]

It's tempting to look at trends and assume there must be a rule behind them, but it's also intellectually lazy. Please do the hard work of justifying your stance like GGP did.

[bdangubic]

it is a simple stance - if you have a product that is used by hundreds of millions of people ad monetization strategy will be found cause there are people a lot smarter than you and me that will get it done. here’s intellectual challenge - find a business with comparable number of users to openai which is not swimming in ad revenue - one will do

[adventured]

For several early years search was thought to have no great business model (banner ads and similar). And then it did.

GoTo.com -> Google -> $$$

[bandrami]

It will not be lucrative if the hundreds of billions of dollars in ad revenue cost the providers trillions in compute. That's still the big question.

[aworks]

I think this is the right question. My search/chat is now 50% ChatGPT, 30% Google, and 20% Gemini. I have no idea the business implications of this.

For that matter, I don't know the bandwidth and compute cycle tradeoffs between traditional search and AI.

[steveBK123]

> Just take a look at the enterprise. Amazon's ad business alone is already a better business than Oracle or SAP or Salesforce, with superior margins, and it's growing faster too.

You can say the same about AWS and then prove the b2b case instead of ad case as well

[adventured]

AWS is legitimately a giant and it should be considered in enterprise broadly. It's infrastructure more than enterprise software of course, which is where Anthropic is at. Anthropic is not trying to host the world's databases and services (at present anyway). Anthropic will however help you write software to compete with Salesforce, Oracle, SAP, et al.

Google's ad business remains far larger and more profitable than AWS. And the advertising segment is drastically larger than the segment AWS is in. Just Google + Meta = nearing $600 billion in ad sales. Amazon will soon have their own $100 billion in ad sales.

[steveBK123]

I guess the question is how many more $100B of ad sales slots are available, aside from just stealing share from incumbents (who already took it from traditional media channels over last 20 years).

At some point someone needs to add value to the real economy, not just take an ad tax off the top.

[bombcar]

I don't think advertising is entirely a zero-sum game (there is real value to everyone for steveBK123 to learn of a product that actually solves a real need or saves money, etc) - but it has to be something akin to it - the economy can't support five hundred trillion dollars in advertising spend.

[iAMkenough]

Not interested in a service with ads throughout my workday, which is why I switched to Anthropic.

Billions in projected revenue is nothing but hype/cope. Google and Meta got their edge because their product was offered for "free" to the masses.

[bdangubic]

absolutely not the case. there isn’t a single nerve in human brains that goes “oh imma tolerate ads cause this shit’s free but if I pay a few bucks no way” - if the product you use has utility to you, you will tolerate ads provided no other acceptable alternative. not to tell you something you don’t already know but anthropic is getting ads, eventually, it is a given. so while today you may have an alternative (arguably better even if no ads in the equation) at some point you won’t have an alternative (other than running local) and you’ll tolerate ads. the thing with LLM ads is that companies can make $$$$ from “ads” you don’t see, i.e. I can (not now but in the future) companies to push my product, e.g. claude is setting up architecture and proposes upstash (which I own and am paying anthropic a lot of money) instead of any competitor. or even more silently adding dependencies on my NPM library which has free and commercial offering…

[iAMkenough]

Yeah sure, but for me the common man OpenAI doesn't add any value that Claude, Gemini or Meta AI doesn't also provide.

If they want to out-ad those companies to the tune of billions, I'll go with the least annoying. OpenAI hasn't earned any loyalty.

[bdangubic]

to me and you sure, but what do you reckon how many of their MAU are just people for whom “AI” is ChaptGPT? 90+%?

[zozbot234]

Anthropic is already an ad - for Anthropic. The uniquely "friendly" personality of their user-facing AI chatbot is far from accidental.

[bdangubic]

hm not sure I follow - the anthropic is selling itself to users that are already using it by being friendly to the users?

[Razengan]

Holy f'n Hell, there's such a blatant bias on HackerNews in favor of Anthropic and against OpenAI.

I'm just a user, and in my experience Claude has been consistently crap compared to ChatGPT/Codex.

I use both side-by-side, and have paid for a ChatGPT subscription every month for around 1 year, but only 2 months for Claude; once last year, and again since last month.

Everything from the sign up, the sign in, the payment, the UI, the UX, gosh, just sucks on Claude.

And the AI itself: SO. MUCH. "OoPs you're right! I was mistaken" BACKTRACKING! It's downright DANGEROUS to listen to it! God I can post screenshots of working on the same project and the same prompts with both agents and prove how worse Claude is.

Of course this comment will be downvoted by Anthropic's paid PR machine, because there's no way actual users who have tried both products would be so in favor of Claude.

[p1esk]

I’ve been a paid subscriber for all three players since day 1. CC (Opus) has been a clear winner for agentic coding starting about 6 months ago. GPT5.4 reduced the gap somewhat but the gap is still there.

[jmye]

> Of course this comment will be downvoted by Anthropic's paid PR machine, because there's no way actual users who have tried both products would be so in favor of Claude.

Sure, it couldn't possibly be that others have had a different experience. It couldn't even be that some people think OpenAI is nearly as gross as Palantir. It's that they're shills.

High-end analysis.

[greenchair]

Yes I dont get it either. probably bots, paid posts.

[paxys]

If/when the bubble bursts Anthropic is going down as well. There's nothing unique that sets it apart from OpenAI. Their cash burn is similarly egregious.

[therobots927]

Relative to OAI they are healthier.

That isn’t saying much.

[sharadov]

The circular deals are getting old - it's like rearranging the chairs on the Titanic's deck.

[xnx]

OpenAI != The AI industry

OpenAI has stagnated technologically, and is a financial zombie, but that's not true for every part of the industry. Once these early movers flame out, there will be more stability with Google, Microsoft, and AWS.

[therobots927]

All the “raises” consist of “committed capital” and all of the revenue is annualized.

Welcome to dot com 2.0

[nacozarina]

sell the roadmap, deliver a sku, sell/comp consulting services on escalations

the silicon valley shuffle, tried & true