[michaeldoron]

Every time an analyst gives the current state of AI-based tools as evidence supporting AI disruption being just a hype, I think of skeptics who dismissed the exponential growth of covid19 cases due to their initial low numbers.

Putting that aside, how is this article called an analysis and not an opinion piece? The only analysis done here is asking a labor economist what conditions would allow this claim to hold, and giving an alternative, already circulated theory that AI companies CEOs are creating a false hype. The author even uses everyday language like "Yeaaahhh. So, this is kind of Anthropic’s whole ~thing.~ ".

Is this really the level of analysis CNN has to offer on this topic?

They could have sketched the growth in foundation model capabilities vs. finite resources such as data, compute and hardware. They could have wrote about the current VC market and the need for companies to show results and not promises. They could have even wrote about the giant biotech industry, and its struggle with incorporating novel exciting drug discovery tools with slow moving FDA approvals. None of this was done here.

[Terr_]

> I think of skeptics who dismissed the exponential growth of covid19 cases due to their initial low numbers.

Compare: "Whenever I think of skeptics dismissing completely novel and unprecedented outcomes occurring by mechanisms we can't clearly identify or prove (will) exist... I think of skeptics who dismissed an outcome that had literally hundreds of well-studied historical precedents using proven processes."

You're right that humans don't have a good intuition for non-linear growth, but that common thread doesn't heal over those other differences.

[actuallyalys]

Yeah, for this analogy to work, we’d have to see AI causing a small but consistently doubling amount of lost jobs.

[mitthrowaway2]

If that were happening right now, how would we know? COVID-19 cases were tracked imperfectly but pretty well; is there any equivalent for AI-related job losses?

[actuallyalys]

Right, my point is that we don't have the data to make a similar exponential argument. We can't rule out the possibility that we're currently in the early stages of exponential growth based on direct measurement. If it is exponential, once it doubles enough times, it will show up in overall economic data.

We can also look at the tools, which have improved relatively quickly but don't appear to be improving exponentially. GPT-4 and GPT-4o came out about a year after their predecessors. Is GPT-4o a bigger leap that GPT-4 was? Are GPT-4.5 or 4.1 a bigger leap than GPT-4 was? I honestly don't know, but the general reception suggests otherwise. The biggest leaps recently seem to be making models that perform roughly as well as past ones but are much smaller. That has advantages from the standpoint of democratization and energy consumption, but those kinds of improvements seems to favor a situation where AI augments workers rather than replaces them.

[bgwalter]

Why not use the promised exponential growth of home ownership that led to the catastrophic real estate bubble that burst in 2008 as an example?

We are still dealing with the aftereffects, which led to the elimination of any working class representation in politics and suppression of real protests like Occupy Wall Street.

When this bubble bursts, the IT industry will collapse for some years like in 2000.

[michaeldoron]

The growth of home ownership was an indicator of real estate investment, not of real world capabilities - once the value of real estate dropped and the bubble burst, those investments were worth less than before, causing the crisis. In contrast, the growth in this scenario is the capabilities of foundation models (and to a lesser extent, the technologies that stem out of these capabilities). This is not a promise or an investment, it's not an indication of speculative trust in this technology, it is a non-decreasing function indicating a real increase in performance.

[mjburgess]

You can pick and choose problems from history where folk belief was wrong: WW1 vs. Y2K.

This isn't very informative. Indeed, engaging in this argument-by-analoguy betrays a lack of actual analysis, credible evidence and justification for a position. Arguing "by analogy" in this way, which picks and chooses an analogy, just restates your position -- it doesnt give anyone reasons to believe it.

[TheOtherHobbes]

I'm not seeing how comparing AI to a virus that killed millions and left tens of millions crippled is an effective way to support your argument.

[drewcon]

Humans are not familiar with exponential change so they have almost no ability to manage through exponential change.

Its an apt comparison. The criticisms in the cnn article are already out date in many instances.

[bayarearefugee]

As a developer that uses LLMs, I haven't seen any evidence that LLMs or "AI" more broadly are improving exponentially, but I see a lot of people applying a near-religious belief that this is happening or will happen because... actually, I don't know? because Moore's Law was a thing, maybe?

In my experience, for practical usage LLMs aren't even improving linearly at this point as I personally see Claude 3.7 and 4.0 as regressions from 3.5. They might score better on artificial benchmarks but I find them less likely to produce useful work.

[drewcon]

5 years ago commercial image gen produced hallucinatory dream like blobs.

2 years ago it was cool but unreliable.

Today I just did an entire “photo shoot” in Midjourney.

[const_cast]

Viruses spread and propagate themselves, often changing along the way. AI doesn't, and probably shouldn't. I think we've made a few movies on why that's a bad idea.

[geraneum]

> Humans are not familiar with exponential change

Humans are. We have tools to measure exponential growth empirically. It was done for COVID (i.e. epidemiologists do that usually) and is done for economy and other aspects of our life. If there's to be exponential growth, we should be able to put it in numbers. "True me bro" is not a good measure.

Edit: typo

[margalabargala]

There's individual persons modelling exponential change just fine, and then there's what happens when you apply to the populace at large.

"A person is smart. People are dumb, panicky dangerous animals and you know it."

[geraneum]

> when you apply to the populace at large

What does this mean? What do you apply to populace at large? Do you mean a populace doesn’t model the exponential change right?

[margalabargala]

Yep that's what I meant! Context clues did you well here.

[agarren]

> The criticisms in the cnn article are already out date in many instances.

Which ones, specifically? I’m genuinely curious. The ones about “[an] unfalsifiable disease-free utopia”? The one from a labor economist basically equating Amodei’s high-unemployment/strong economy claims to pure fantasy? The fact that nothing Amodei said was cited or is substantiated in any meaningful way? Maybe the one where she points out that Amodei is fundamentally a sales guy, and that Anthropic is making the rounds saying scary stuff just after they released a new model - a techbro marketing push?

I like anthropic. They make a great product. Shame about their CEO - just another techbro pumping his scheme.

[dingnuts]

especially when the world population is billions and at the beginning we were worried about double digit IFR.

Yeah. Imagine if COVID had actually killed 10% of the world population. Killing millions sucks, but mosquitos regularly do that too, and so does tuberculosis, and we don't shut down everything. Could've been close to a billion. Or more. Could've been so much worse.

[IshKebab]

I think you missed the point. AI is dismissed by idiots because they are looking at its state now, not what it will be in future. The same was true in the pandemic.

[monkeyelite]

> I think of skeptics who dismissed the exponential growth of covid19 cases due to their initial low numbers.

But that didn’t happen. All of the people like pg who drew these accelerating graphs were wrong.

In fact, I think just about every commenter on COVID was wrong about what would happen in the early months regardless of political angle.

[tim333]

I remember scientists, especially epidemiologists being quite accurate. I guess the key is to not even have a political angle but instead some knowledge of what you are talking about.

[monkeyelite]

I don’t remember that. How do you explain the total collapse of public trust in science and medical institutions?

Try revisiting their content from spring of 2020 (flatten the curve, wild death predictions, etc).

> I guess the key is to not even have a political angle

It’s a fantasy to imagine technical knowledge allows you to transcend the political and 2020 only reinforced that.

[timr]

> I think of skeptics who dismissed the exponential growth of covid19 cases due to their initial low numbers.

Uh, not to be petty, but the growth was not exponential — neither in retrospect, nor given what was knowable at any point in time. About the most aggressive, correct thing you could’ve said at the time was “sigmoid growth”, but even that was basically wrong.

If that’s your example, it’s inadvertently an argument for the other side of the debate: people say lots of silly, unfounded things at Peak Hype that sound superficially correct and/or “smart”, but fail to survive a round of critical reasoning. I have no doubt we’ll look back on this period of time and find something similar.

[SoftTalker]

Analysis == Opinion when it comes to mainstream news reporting. It's one guy's thinking on something.

[deadbabe]

It goes both ways. Once the exponential growth of COVID started, I heard wildly outrageous predictions of what was going to happen next, none of which ever really came to fruition.

[biophysboy]

Its an article reformulated from a daily newsletter. Newsletters take the form of a quick, casual follow up to current events (e.g. an Amodei interview). Its not intended to be exhaustive analysis.

Besides the labor economist bit, it also makes the correct point that tech people regularly exaggerate and lie. A great example of this is biotech, a field I work in.

[qgin]

This is the exact thing I’ve expressed as well.

This moment feels exactly to me like that moment when we were going to “shut down for two weeks” and the majority of people seemed to think that would be the end of it.

It was clear where the trend was going, but exponentials always seem ridiculous on an intuitive level.

[PeterStuer]

"Is this really the level of analysis CNN has to offer on this topic?"

It's not CNN exlusive. Newsmedia that did not evolve towards clicks, riling up people, hatewatching and paid propaganda to the highest bidder went extinct a decade ago. This is what did evolve.

[biophysboy]

This is outdated. Most of journalism has shifted to subscription models, offering a variety of products under one roof: articles, podcasts, newsletters, games, recipes, product reviews, etc.

[aaronbaugher]

> Is this really the level of analysis CNN has to offer on this topic?

Not just this topic.

[bckr]

That’s not what major news outlets are for. I’m not sure exactly what they’re for.

[leeroihe]

The best heuristic is what people are realizing happened with uncheck "skilled" immigration in places like canada (and soon the U.S.). Everyone was sold that we "need these workers" because nobody was willing to work and that they added to GDP. When in reality, there's now significant evidence that all these new arrivals did was put a net drain on welfare, devalue the labor of endemic citizens (regardless of race - in many cases affecting endemic minorities MORE) and in the end, just reduced cost while degrading companies who did this.

We will wake up in 5 yrs to find we replaced people for a dependence on a handful of companies that serve llms and make inference chips. Its beyond dystopian.

[matteotom]

Can you provide more details about said "significant evidence"? This seems to be a pretty popular belief, despite being contrary to generally accepted economics, and I've yet to see good evidence for it.