What if A.I. doesn't get better than this? | Hacker News Reader

What if A.I. doesn't get better than this?

newyorker.com

90 pointssundache112 comments

My current intuition on this topic is that they are right about scaling but they are training on the wrong data.

LLMs were not intended to be the core foundation of artificial intelligence but an experiment around deep learning and language. Its success was an almost accidental byproduct of the availability of large amount of structured data to train from and the natural human bias to be tricked by language (Eliza effect).

But human language itself is quite weak from a cognitive perspective and we end up with an extremely broad but shallow and brittle model. The recent and extremely costly attempts to build reasoning around don't seem much more promising than using a lot of hardcoded heuristics, basically ignoring the bitter lesson.

I've seen many argue that a real human level AI should be trained from real-world experience, I am not sure this is true, but training should likely start from lower-level data than language, still using tokens and huge scale, and probably deeper networks.

Not all AI is LLMs. That's just what's most prevalent right now. There's still great work being done by models that don't "speak" but "perform". The issue is they need to be trained to perform like you said. The more tools like Claude Code are used, the more training they receive as well. I do think we'll see a plateau (if we haven't reached it already) of diminishing returns and we'll seek out new algorithms to improve it.

Never underestimate the will of someone determined to gain an extra 10% performance or accuracy. It's the last 1% I worry about. 99.99% uptime is great until it isn't. 99% accuracy is great until it isn't. These things could be mitigated by running inference on different quantinizations of a model tree but ultimately we're going to have to triple check the work somehow.

> The more tools like Claude Code are used, the more training they receive as well.

What do you mean? A model doesn't improve because it's being used more. Are you saying Anthropic invests more into Claude Code the more people use it? Or are you saying they collect its output and train it on it?

I assume they mean that they can gather users inputs (e.g. the user correcting the model, suggesting improvements, etc.).

They probably mean https://en.wikipedia.org/wiki/Reinforcement_learning_from_hu..., but I don't think that is a huge factor for Claude Code.

Definitely smarter people than me have thought about this already, but I’ve been trying to think about human language and how thoughts form in my head lately. How does thinking feel to you?

I feel like thoughts appear in my head conceptually mostly formed, but then I start sequentially coming up with sentences to express them, almost as if I’m writing them down for somebody else. In that process, I edit a bunch, so the final thought is influenced quite a bit by how English ends to be written. Maybe even constrained by expressability in English. But English has the ability to express fuzzy concepts. And the kernel started as a more intuitive thing.

It is a weird interplay.

LLMs aren't monolingual so you might want to expand that beyond English. Consider how multilingual people think.

Also, apparently it's pretty common for people to think in words and have an internal monologue. I hadn't realized this was a thing until recently but it seems many people don't think abstractly as you've described.

There have been psychological experiments that have shown what OP experiences.

In this particular experiment subjects were asked to make a series of arbitrary choices between 2 items, while being hooked up to some brain scanning equipment. Subjects were asked to think about it for 5 seconds before making the choice.

The experiment showed that scientists were able to predict the choice before the subjects consciously reasoned about it. So the experiment indicates that the choice is made subconsciously, and the reasoning ends up on whatever choice was made.

This aligns pretty well with some other psychological theories, and it points at that a lot/most of our brain processes happen subconsciously, and our conscious experience mainly serves as a way of providing a coherent story about why we experience what we experience.

So it seems that no matter if you 'think' abstractly or visually or with a monologue, this is merely a small step in our overall cognition, and doesn't really change the fact that thing bubble up from our subconscious before we become conscious of it.

5 seconds is an interesting amount of time. I think we all can simulate an argument in our heads, and it could even convince us. But, that takes time. 5 seconds seems like just enough time to result a thought into words.

I wonder if they could plot the mind-changing process resulting from that sort of argument. If that is what even happens.

Since you understand predictive coding, I believe you will be interested in this:

https://github.com/dmf-archive/IPWT

What happens is they go out of business: "these firms spent five hundred and sixty billion dollars on A.I.-related capital expenditures in the past eighteen months, while their A.I. revenues were only about thirty-five billion."

DeepSeek (and the like) will prevent the kind of price increases necessary for them to pay back hundreds of billions of dollars already spent, much less pay for more. If they don't find a way to make LLMs do significantly more than they do thus far, and a market willing to pay hundreds of billions of dollars for them to do it, and some kind of "moat" to prevent DeepSeek and the like from undercutting them, they will collapse under the weight of their own expenses.

DeepSeek is also undercutting itself. No one is making a profit here, everyone is trying to gobble market share. Even if you have the best model and don't care to make a dime, inference is very expensive.

I'd be surprised if Google weren't closer to profitability than basically anyone else, as they have their own hardware and have been running these kinds of applications for much longer than anyone else.

Google is still investing heavily in data centres. Presumably without the AI they could lift their foot off that throttle.

Google are well set up to monetize compared to the others. They just need people to see their ads.

They are also the ones with more to lose to LLMs

DeepSeek doesn’t need to make a profit to be successful.

I could think of several reasons this could be so, but it would be good to hear your logic on this claim?

State funded innit

> inference is very expensive

I am surprised that this claim keeps getting made, given the observed prices.

Even if one thinks that the losses of big model providers are due to selling below operating costs (rather than below that plus training costs plus the cost of growth), then even big open-weights models that need beefy machines, look like they eventually* amortise the cost so low that electricity is what matters; so when (and *only* when) the quality is good enough, inference is cheaper than the food needed to have a human work for peanuts — and I mean literally peanuts, not metaphorical peanuts, as in the calories and protein content of bags of peanuts sufficient to not die.

* this would not happen if computers were still following the improvements trends of the 90s, because then we'd be replacing them every few years; a £10k machine that you replace every 3 years cost you £9.13/day even if it did nothing.

https://www.tesco.com/groceries/en-GB/products/300283810 -> £0.59 per bag * (2500 per day/645 per bag) = £2.29/day; then combine your pick about which model, which model of home server, electricity costs etc. with your estimate of how many useful tokens a human does in 8,760 hours per calendar year given your assumptions about hours per working week and days of holiday or sick leave.

I know that even just order-of 100k useful tokens is implausible for any human because that would be like writing a novel a day, every day; and this article (https://aichatonline.org/blog-lets-run-openai-gptoss-officia...) claims a Mac Studio can output 65.9/second = 65.9 * 3600 * 24 = 5,693,760 / day or ~= 2e9/year, compare to a deliberate over-estimate of human output (100k/day * 5 days a week * 47 weeks a year = 2.35e7/year)

The top-end Mac Studio has a maximum power draw of 270 W: https://support.apple.com/en-us/102027

270 W for *at least (2e9/year / 2.35e7/year) 85 times* the quantity (this only matters when the quality is sufficient, and as we all know AI often isn't that good yet) of output that a human can do with 100 W, is a bit over 31 times the raw energy efficiency, and electricity is much cheaper than calories — cheaper food than peanuts could get the cost of the human down to perhaps about £1/day, but even £1/day is equivalent to electricity costing £1/(24 hours * 100 W) = £0.416666… / kWh

Running a local model is not an apples comparison. Yes, if you run a small model 24/7 without a care for output latency and utilization is completely static with no bursts, then it can look cheap. But most people want output now, not in 10 hours. And they want it from the best models. And they want large context windows. And when you combine that with serving millions of users, it gets complicated and expensive.

When you combine that with serving millions of users, it also gets amortised over several million users.

> But most people want output now, not in 10 hours.

At 65t/s, that's 2.5 million tokens output.

Yes, but usage is not uniform even when you have millions of users. It smooths the usage lines, but the peaks and troughs become more extreme the more users you have. At 3am usage in the US goes down to effectively 0. Maybe you can use the compute for Asia customers, but then you compete with local compute that has far better latency.

Then you have seasonal peaks/troughs, such as the school year vs summer.

When you want 4 9s of uptime and good latency, you either have to overprovision hardware and eat idling costs, or rent compute and pay overhead. Both cost a lot.

I'm not familiar with the actual strategy, but one strategy of something like DeepSeek or any open model could be, roughly, to avoid moat creation by other companies. E.g. one reason Google maintained dominance for so long is they simply paid off every smart person and company that might build competitors.

Matt Levine of Bloomberg said he believes most big hedge funds have a proprietary LLM (and various other machine learning models), which they use for their own purposes anyway, and therefore it was relatively straightforward for them to get into the business.

> If they don't find a way to make LLMs do significantly more than they do thus far...
They only need two things, really: A large user base and a way to include advertising in the responses. The market willing to pay hundreds of billions of dollars will soon follow.
The businesses are currently in the user base building stage. Hemorrhaging money to get them is simply the cost of doing business. Once they feel that is stable, adding advertising is relatively easy.

> and some kind of "moat" to prevent DeepSeek and the like from undercutting them*

Once users are accustomed to using a service, you have to do some pretty horrendous things to get them to leave. "Give me your best hamburger recipe" -> "Sure, here is my best burger recipe [...] However, if you don't feel like cooking tonight, give the Big Mac a try!". wouldn't be enough to see any meaningful loss of users.

That's nothing new. The question is, will those users be willing and able to pay ten times as much or more for those same services they get for a significant discount now.

Will they pay ten times more for a Big Mac? Probably not, but why would they need to? Hundreds of billions is Facebook's revenue. The businesses in this space are there if they can take those customers alone, never mind all the other places where advertising takes place. The market exists, is sufficiently large, and willing to spend. All these "AI" businesses need to do is show that the users are spending time on their services instead, which is exactly what they are working on right now.

The question is really only: Will users actually want to continue to use these services once the novelty wears off? The assumption is that they are useful enough to become an integral part of our lives, but time will tell...

> The question is really only: Will users actually want to continue to use these services once the novelty wears off? The assumption is that they are useful enough to become an integral part of our lives, but time will tell...

But LLMs do have some niche, stable applications already. For example, they replaced Stack Overflow to a large extent, because you can get the answer you need faster, and it's often better adapted to your situation. So you could argue the novelty of SO wore off a long time ago but people were still using it when LLMs appeared. ChatGPT is no more en vogue, people are ashamed to (and shamed for) using it, but it still has some uses, helping people in their jobs and lives in general.

Stack Overflow wasn't bringing in hundreds of billions in revenue and never could have hoped to. A niche won't cut it.

I mean, should it all come crashing down and once the dust settles there is no doubt room for a niche service to rise from the ashes. Many have predicted exactly that AI will have its own "dark fibre" moment. But the current crop of providers seeking "world domination" won't survive if they can only carve out a SO-style niche.

> Stack Overflow wasn't bringing in hundreds of billions in revenue

Neither do LLMs. Instead, they cost hundreds of billions. Whether those investments can ever be recovered is still an unanswered question.

> Neither do LLMs.

Not yet. Of course, if you had read the thread you'd know that LLM businesses are trying to see if they can capture Facebook-scale/beyond user bases in order to serve ads to them at Facebook-scale/beyond.

> Whether those investments can ever be recovered is still an unanswered question.

Yes, that's the question we are discussing. Welcome to many comments ago.

The world is not a zero-sum game, but I don't think it's likely that companies will double their advertising. So either Google, Meta, etc. will lose the AI game and most of the advertising revenue will shift to different companies or Google, Meta, etc. win and they get to keep their advertising income.

I don't see how suddenly hundreds of billions of additional ad revenue will appear.

I think some of AI companies truly wanted to make a fortune replacing all white collar workers through model exclusivity, but that open models (initially Llama and then a sequence of really good Chinese models) threw a wrench in the cogs. There is not as much you can make anymore if everyone can host their own 'workers' near cost price.

> So either Google, Meta, etc. will lose the AI game

They might lose entirely (search, social media, etc.) if users are more likely to direct their eyeballs to "AI" services. And that isn't an impossible scenario. What do you need traditional web search, stupid internet comments, etc. for when an LLM will generate all that and more immediately at your behest? The newspaper companies lost when other forms of media came along. "AI" could easily push things the same way.

But the businesses are still trying to prove that. It is still early days. Only time will tell if they will actually get the aforementioned user base.

They might lose entirely (search, social media, etc.) if users are more likely to direct their eyeballs to "AI" services.

Sorry that I wasn't clear enough. That was what I was trying to imply - in that case the ad purchases that currently go to Google etc. will go to a new AI company. This is why both Google is investing so heavily in LLMs, they know that LLMs can kill search + website visits and thus cause them to lose their ad income.

But either way, it's questionable if it will increase all ad spending. So either AI companies collapse and a lot of VC money is burned or Google et al. collapse and a lot capital is burned (pension funds and whatnot).

Why would ad spending need to increase? Google, Meta, et al. collapsing and an "AI" business taking their place is fine. The universe doesn't care.

This is why Google has android and chrome, and why Facebook bought WhatsApp and Instagram. It isn’t so much about spending billions on growth, it is about hedging their bets and protecting their monopolistic cash cows.

I don’t see how advertising is going to work with agents, especially if they’re being used by companies to replace or supplement jobs. Am I going to have comments in my code with ads for McDonalds? Will the AI support agent start trying to sell me a VPN?

I don’t think any of these AI companies can justify their expenses without meaningfully automating a significant amount of white collar work, which is yet to happen.

> I don’t think any of these AI companies can justify their expenses without meaningfully automating a significant amount of white collar work

The businesses in question are building chatbots, not trying to automate jobs. Their primary goal at this stage is to get legions of people hooked on chatting with their service on a regular basis. Once (if) they have succeeded with that, then they can move on to step two.

> Am I going to have comments in my code with ads for McDonalds?

"Give me a function which stores a supplied string to a file in $HOME." -> "I have updated your code with said function. It [...] Have you considered storing the file on Amazon S3 for additional robustness and reliability?"

We did a test of GPT5 yesterday. We asked it to generate a synopsis of a scientific topic and cite sources. We then checked those sources. GPT5 still hallucinated 65% of the citations. It did things like: Make up the paper title Make up the authors for a real paper title Mix a real title and a real journal If it can't even reference real papers it certainly can't be trusted to match up claims of fact with real sources.

Current AI tools generate citations that LOOK real but ARE fake. This might not be solvable inside the LLM. If anyone could do it, it'd be OpenAI. (OK maybe I'm giving them too much credit, but they have a crap-ton of money and seem to show a real interest in making their AI better)

If it can't be done in the LLM we can't trust LLMs basically ever. I suppose there's a pretty big loophole here. Doing it outside the LLM but INSIDE the LLM product would be good enough.

The first AI tool to incorporate that (internal citation and claim checking) will win because if the AI can check itself and prevent hallucinated garbage from ever reaching the user we can start to trust them and then they can do everything we've been promised. Until that day comes we can't trust them for anything.

Google already did this, give free gemini deepresearch a spin. It's not perfect, but I have a feeling you'll be surprised if this is your honest impression.

The title is irritating, conflating AI with LLMs. LLMs are a subset of AI. I expect future systems will be mobs of expert AI agents rather than relying on LLMs to do everything. An LLM will likely be in the mix for at least the natural language processing but I wouldn't bet the farm on them alone.

The computing power alone of all these gpus would bring a revolution in simulation software. I mean 0 AI/machine-learning, just being able to simulate much more things than we can.

Most industry-specific simulation software is REALLY crap, most from the 90s and 80s and barely evolved since then. Many stuck on single core CPUs.

It could be a nice side-effect of having all this “LLM hardware” built into everything, nice little throughput focused accelerators in everybody’s computers.

I think if I were starting grad school now and wanted some easy points, I’d be looking at mixed precision numerical algorithms. Either coming up with new ones, or applying them in the sciences.

That battle was long-ago lost when the leading LLM companies and organizations insisted on referring to their products and models solely as "AI", not the more-specific "LLMs". Implementers of that technology followed suit, and that's just what it means now.

You can't blame the New Yorker for using the term in its modern, common parlance.

Agreed, and ultimately it's fine because they're talking about products not technology. If these products go in a completely different direction and LLMs become obsolete the AI label will adapt just fine. Once these things hit common parlance there's no point in arguing technical specificity as 99.99% of the people using the term don't care, will never care, and language will follow their usage not the angry pedant.

Sure I can. If someone writing for the New Yorker has conflated the two concepts and is drawing bad conclusions because of it, that’s bad writing.

A good writer would tease apart this difference. That’s literally what good writing is about: giving a deeper understanding than a lay person would have.

This is something I immediately noticed when ChatGPT first released. It was instantly called "AI", but previous to that, HN would have been up in arms that it's "machine learning" not actual intelligence. For some reason, the crowd here and everywhere else just accepted the misuse of the word intelligence and let it happen. Elsewhere I can understand, but people here know better.

Intentionally misconstruing it as actual intelligence was all a part of the grift from the beginning. They've always known there's no intelligence behind the scenes, but pushing this lie has allowed them to take in hundreds of billions in investor money. Perhaps the biggest grift the world has ever seen.

A* search, literally textbook AI, is still doing great work.

If the New Yorker published a story titled "What if LLMs Don't Get Better Than This?" I expect the portion of their readers who understood what that title meant would be pretty tiny.

Indeed, and the title itself contains an operational definition of AI as “this” (LLM’s) - if AI becomes more than “this” then the question has been answered in the affirmative.

AI is LLMs now. Similar to how machine learning became AI 5-10 years ago.

I'm not endorsing this, just stating an observation.

I do a lot of deep learning for computer vision, which became AI a while ago. Now, when you use the word AI in this context, it will confuse people because it doesn't involve LLMs.

AI is what people think AI is. In the 80s, that was expert systems. In the 2000s, it was machine learning (not expert systems). Now, it is LLMs — not machine learning.

You can complain, but it’s like that old man shaking their fist at the clouds.

Now, if you want to talk about cybernetics…

The title annoys me more because if doesn't mention anything about time. AI will almost certainly get a good bit better eventually. The questions will it in the next couple of years or will we have to wait for some breakthrough.

I'm amused they seem to refer to Marcus and Zitron as "these moderate views of A.I". They are both pretty much professional skeptics who seem to fill their days writing AI is rubbish articles.

> You didn’t need a bar chart to recognize that GPT-4 had leaped ahead of anything that had come before.

You did though. I remember when GPT-4 was announced, OpenAI downplayed it and Altman said the difference was subtle and wouldn't be immediately apparent. For a lot of the stuff ChatGPT was being used for the gap between 3 and 4 wasn't going to really leap out at you.

https://fortune.com/2023/03/14/openai-releases-gpt-4-improve...

In the lead up to the announcement, Altman has set the bar low by suggesting people will be disappointed and telling his Twitter followers that “we really appreciate feedback on its shortcomings.”

OpenAI described the distinction between GPT-3.5—the previous version of the technology—and GPT 4, as subtle in situations when users are having a “casual conversation” with the technology. “The difference comes out when the complexity of the task reaches a sufficient threshold—GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5,” a research blog post read.

In the years since we got a lot more demanding of our models. Back then people were happy if they got models to write a small simple function and it worked. Now they expect models to manipulate large production codebases and get it right first time. So, the difference between GPT-3 and GPT-4 would be more apparent. But at the time, the reaction was somewhat muted.

> Back then people were happy if they got models to write a small simple function and it worked. Now they expect models to manipulate large production codebases and get it right first time.

This push is mostly coming from the C-level and the hustler types, both of which need this to work out in order for their employeeless corporation fantasy to work out.

I'm not going to say that nobody is expecting it to do these things, but I don't think they should. It's still unable to write a simple function.

What we've seen isn't a reasonable increase in expectations based upon validation of previous experiments. Instead it's racking up of expectations by all the signals of success. When they time and time again take in more VC cash at ever greater valuations, we are forced to assume they want to do something more, and since they get the cash we have to assume somebody believes them.

Its a pyramid scheme, but instead of paying out earlier investors with the later investors cash its a confidence pyramid scheme. They obsolete the previous investors valuations by making bigger claims with larger expectations. Then they use those larger expectations as proof they already fulfilled the previous expectations.

> This push is mostly coming from the C-level and the hustler types, both of which need this to work out in order for their employeeless corporation fantasy to work out.

The irony, at least in my mind, is that C-level hustler types are exactly the perfect role to be replaced by "AI" for big cost-savings. For obvious reasons, it won't happen.

Frankly, I'm not sure if LLM would be the best replacement. Given they need to make decisions, I'd say ensemble methods could work better, at least in scenarios where you can actually perform reliable simulations (e.g. production, but not necessarily marketing). LLMs tend to averages whereas what shareholders want is to optimize for X, where X is usually, but not necessarily, net profit.

(It would be a fun experiment to create such an optimization startup and advertise it to boards as "AI-CEO-as-a-service" and watch the faces of these CEOs pushing for AI in workplace. Maybe they would start having a more nuanced view.)

There's an entire category of things designed for game theory and other types of zero sum games like chess, go, poker, etc so I don't see why you can't have an "ai" system that's making profit-orientated, automated decisions.

OpenAi has 700+ million users. Sam recently said only 7% of Plus users were using thinking (o3)!!! That means 93% of their users were using nothing but 4o!

Clearly the OpenAi leadership saw these stats and understood the main initial goal of GPT5 is to introduce this auto-router, and not go all in on intelligence for the 3-7% who care to use it.

This is a genius move IMO, and will get tons of users to flood to ChatGPT over competitors. Grok, Gemini, etc are now fighting over scraps of the top 1% while OpenAi is going after the blue ocean of users.

If they're not paying users then they're just a liability.

monetizing those will come eventually - it's just hard to get right

> (monetizing LLMs is) just hard to get right

Time will tell if that's just a euphemism for "there is no business model here".

How so? You capture the market first, then you turn on paid ads and reap benefits for decades like Google.

Even when a business makes a profit, then the costs are still technically liabilities.

And you're assuming that ad revenues or whatever will be high enough to cover the costs. And that's not a given.

> Sam recently said only 7% of Plus users were using thinking (o3)

Thinking or just o3, and over what timeframe? There were a lot of days where I would just rely on o4-mini and o4-mini (high) b.c. my queries weren't that complex and I wanted to save my o3 quota and get faster responses.

> That means 93% of their users were using nothing but 4o!

Also potentially 4.1 and 4.5?

> the percentage of users using reasoning models each day is significantly increasing; for example, for free users we went from <1% to 7%, and for plus users from 7% to 24%.

https://x.com/sama/status/1954603417252532479

How can you say progress has stalled two weeks after LLMs won gold medals at IOI and IMO?

How can you say progress has stalled without having visibility on the compute costs of gpt-5 relative to o3?

How can you say progress has stalled by referring to changes in benchmarks at the frontier over just 3.5 months?

All that feels like specialized stunts like IBM’s Watson beating Ken Jennings at Jeopardy.

The rate of improvement has slowed significantly. And chasing benchmarks is making everything worse IMO. Opus 4.1 is worse than Sonnet 3.7 to me :/.

I think the future will be:

1. Ads and quantization/routing to chase profits

2. Local models start taking over. New companies will slide in without the huge losses and provide what Claude/OpenAI do today at reasonable margins

3. Apple/Google eat up lots of the market by shipping good-enough models with iOS/Android

You can't say that with any certainty, but I personally share the impression that growth has not kept up with the hype of 2023. Take the following for example. That's an article from April 2023, that strongly implies that the next version of GPT would be so much more powerful than the current one that it would be dangerous to work on or even release.

Altman specifically used the version number "GPT5" back then. GPT5 is quite good, but is it the kind of technology that requires a word-wide moratorium on its development, lest it make humanity redundant?

"""

(Friedman) asked Altman for his thoughts on the recently released and widely circulated open letter demanding an AI pause. In response, the OpenAI founder shared some of his critiques. “An earlier version of the letter claimed OpenAI is training GPT-5 right now. We are not, and won’t for some time,” Altman noted. “So in that sense, [the letter] was sort of silly.”

But, GPT-5 or not, Altman’s statement isn’t likely to be particularly reassuring to AI’s critiques, as first pointed out in a report from the Verge. The tech founder followed up his “no GPT-5″ announcement by immediately clarifying that upgrades and updates are in the works for GPT-4. There are ways to increase a technologies’ capacity beyond releasing an official, higher-number version of it.

"""

(from: https://gizmodo.com/sam-altman-open-ai-chatbot-gpt4-gpt5-185...)

My personal test question keeps bombing, and I think it's something they should be capable of doing?

Are those math contests? Are their questions and answers in the training set?

Let's say that these things really won a math Olympiad by thinking. Ok, I would like it to to write parsers based on a well defined expression or language spec. Not as bad as near unparseable C++ or JavaScript.

The AIs refuse, despite the prompt, to write a complete parser, hallucinate tests, do things like just call the already working compiler on the CLI, force repetitive reprompts that still won't complete the task.

To me, this is a good example of a task I would give AI as a service to see if it will reliably do something that's well specified, moderately annoying, and is most definitely in the training set if they are pulling data from "the internet".

> My personal test question keeps bombing, and I think it's something they should be capable of doing?

The problem is that "they" isn't a monolith. How much compute went into your tests? Gpt-5 thinking in ChatGPT Plus uses less compute than Gpt-5 thinking in ChatGPT Pro, which uses less compute than the "high" reasoning effort when "gpt-5" is called via the API, which uses less compute than Gpt-5 Pro in ChatGPT Pro, which uses less compute than custom scaffolds, which uses less compute than what went into the IMO/IOI solutions. This is not just my idle speculation, it's publicly available information.

The bear market decade the stock market has been putting off since 2021 with this AI gasping phase happens.

https://www.currentmarketvaluation.com/models/s&p500-mean-re...

https://www.cell.com/fulltext/S0092-8674(00)80089-6

Investors are betting on growth. The public loves the hype. As a user, I already have something useful. Schadenfreude.

AI doesn't need to get better than this. It is already saving millions of hours of previously wasted human productivity. The biggest threat to these companies if their products do not improve is the local running of LLMS. That would finally justify consumers buying more memory and processor speed.

AI getting better is like maybe 50% or less of the equation. The other part is the infrastructure supporting AI applications. The infrastructure and interfaces that need to be built to fully take advantage of whats already here already has a long way to catch up.

It appears that Cal Newport has decided to be the one to most publicly initiate the inevitable Trough Of Disillusionment stage of the Hype Cycle. I'm not sure it'll last very long, though, considering (for starters) Google DeepMind's gold medal at the recent International Math Olympiad. Also, while he criticizes the cost-cutting measure which is ChatGPT 5, he doesn't even mention ChatGPT 5 Pro, which is performing excellently.

I always expect things like this to eventually deliver about 90% of what they promise, which turns out to be 100% for some niche uses, and for the rest it just gets abandoned because 90% isn't good enough. Like when voice recognition became super hyped in the late 90's, it was going to change how the world interacts with machines, and eventually it turned into "Hey Siri"

Good question.

In the one side I read stuff about exponential gains with every new model. On the other side, the coding improvements look logarithmic to me.

Which to me means, what's the Big o of this entire venture?

Ultimately, what they need to do is add nines of reliability. I guess I could argue that what they are producing now is like two nines: 99% accuracy.

Of course, that depends on how you measure it and yada yada yada. So for things like self-driving, I could see how people could argue that the accuracy rate is 99.9% on a minute by minute basis.

But how many nines do you need? Especially for self-driving five more? What's the computational cost to achieve that? Is it just five times? Is it 25 times? Is it two to the five power?

That's completely possible if the development of LLMs follow an S-curve (sigmoidal). At the beginning of the curve it will look exponential, then linear, and finally logarithmic. For different tasks, LLMs could be on different points on the curve, which would explain why some people perceive the improvements as exponential and others perceive them as logarithmic - they are simply working on different things and so experience different gradients.

https://archive.ph/20250813061454/https://www.newyorker.com/...

Thank you for that!

They didn't answer much of the "What if," though... Am just imagining the massive financial losses taken by so many, and if a bailout becomes necessary, because too-big-to-fail now means Microsoft, Google, Facebook et al since we transferred so much of financial engineering economics onto them since '08.

Those three companies have products outside AI and won't die quickly. The ones that will collapse are betting exclusively on improvements in AI. It will be fun to watch the VC money burn.

Last time I've checked each of these companies were still hugely profitable. So it's not going to be your average FANG in trouble here, but rather VCs and others who've jumped onto the AI-craze

I hope the necessary bailout fails due to political fighting in Congress.

LLMs will sure hit a dead end. But I think they will be a major stepping stone to help figure out and write the next generation.

I predict that the article of roughly the same title will be popping up on him every couple of years.

https://web.archive.org/web/20250813134114/https://www.newyo...

AI is so new and so powerful, that we don't really know how to use it yet. The next step is orchestration. LLMs are already powerful but they need to be scaled horizontally. "One shotting" something with a single call to an LLM should never be expected to work. That's not how the human brain works. We iterate, we collaborate with others, we reflect... We've already unlocked the hard and "mysterious" part, now we just need time to orchestrate and network it.

I think you’re right - even if we accept the premise that there’s only room for minor marginal improvements, there’s vast amounts of room for improvement with integrations, mcp, orchestration, prompting, etc. I’m talking mostly about coding agents here but it applies more widely.

It’s a completely new tool, it’s like inventing the internal combustion engine and then going, “well, I guess that’s it, it’s kinda neat I guess.”

I think that's it. Even if there were no improvements with LLMs as they exist today, the integration and usage can still be vastly improved. Right now, we don't have multiple LLM-aware systems communicating, with a standardized information repository.

Right now we have the technology to have an AI observe a room, count the people in it, see what they're doing, observe their mood, and set the lighting to the appropriate level. We just don't have all the sensors and integrations and protocols to manage that. The LLM interfaces with email, your bank, your phone, etc., is crude and clunky. So much more could be done with the LLMs we have now.

(And just to be clear, most of those integrations sound horrible and dystopian. But they're examples.)

Powerful but we don't know how to use it? If it is as powerful as all you true believers spout the usefulness would be self evident and that would be the display of its power.

But apparently it is powerful just because you say so, and then something, something ... business model ...

It does incredible things today that we wouldn't have thought possible 5 years ago.