49 Comments

"Right now, the math for profitability doesn’t work, influence on productivity of GenAI has been minimal, and prospects for pure scaling aren’t looking great."

Yes, it is completely unproductive for all the claimed benefits. However, what makes it worse is that it is actually productive for filling the internet with mindless noise. The scale of content farms using bots is enormous. They said AI would empower the individual. Instead it is empowering those with vast resources to drown out your voice with mass produced hallucinations.

Something I posted about recently here FYI - https://www.mindprison.cc/p/dead-internet-at-scale

Expand full comment

"Gary Marcus apologizes for repeating himself with minor variations. But minor variations seems to be all that is emerging, despite the enormous investments." made be laugh out loud.

I do wonder how you keep it up, though, as having to constantly correct a relentless stream of nonsense/hype is so much negative energy.

Expand full comment

We just have to figure out how to harness negative energy in order to beat them. They, AI, are only able to harness the electric power grid. When/If that shuts down, or gets turned off, all AI will "be no more than dust before the wind."

Expand full comment

Gary, those in the know -- know why it won't. As the data becomes biger and bigger, and bigger still, non ergodic pieces are added to pile. For statistical inference to be reliable, the data need to be ergodic, this is a given. Deep Learning, being based on staistical inference, is able to provide good correlation, Reliability, causality and undertanding is beyond its territoty.

Expand full comment

In my darker moments I wonder if generative AI's lack of economic rationale matters.

Are these digital Potemkin facades being created to turn a profit, or to impose political and social regimes?

There's every likelihood that we're going to hear excuses for chaotic failures that sound like this: "It's the best policy. AI says so."

I have no idea if this will happen, but the odds worry me about my grandkids' future.

Expand full comment

And the responses to the "top 5 things I did last week, sir" email are to be pushed into some AI tool to analyze who's doing worthwhile work.

This is fine.

Expand full comment

I've been very happily retired for some time and don't live in the US. But I got a big laugh imagining what I'd report to those parasites as my "top 5 things" if I was a federal employee 2025. They'd throw me out for sure but I'd be laughing my backside off. 😃

Expand full comment

I'd be worrying about a lot sooner. One doesn't need AI to be more than it already is to wreak havoc on the world, especially in the ways you mention.

Expand full comment

I like you, Gary, you’re like the “grumpy old” man on the porch of the AI neighborhood- just calling it as it unfolds.

Expand full comment

🤣 I wanna be in a Spike Lee movie, playing myself.

Expand full comment

A more honest naming convention would be "Baka-Tensai 1", "Baka-Tensai 2", ..., "Baka-Tensai N". Baka-Tensai means "Stupid-Genius". It's impossible to simply scale the "Stupid" part out.

Expand full comment

Or maybe Half Baka-Tensai

Expand full comment

I constantly have the feeling nobody here is aware of the spectacularly bad “predictions” from Clifford Stoll about “the internet”

https://www.newsweek.com/clifford-stoll-why-web-wont-be-nirvana-185306

As they repeat the same naive statement.

The background use of AI is beginning to become surprising, to me, I have multiple projects now on to deliver things simply not possible for corporate clients as recently as 5 years back, with my first AI project a team delivered in 2019/2020, and is in use.

“Luddites said” nobody would watch TV, nobody would listen to FM radio, nobody would use a telephone, talking movies, home computers, cellphones, 1000 cable channels, VCR, records CD, and apparently nobody is using AI, it’s all pure hype and has no application anywhere for any thing at any time for now or the foreseeable future.

News folks, I used AI two years ago to rewrite and construct large-scale frameworks for process control with large ERP teams, coming online now, more complete and cheaper than I could have believed possible.

The reaper for IT jobs has begun. They will be the gigantic gigantic casualties, since marketing has already been decimated.

Is everyone not aware?

AGI is nonsensical, since we can’t even define it, but we have amply defined neural-network based LLM’s supporting GPT models.

Expand full comment

> The truth is no online database will replace your daily newspaper

Yeah that one didn’t age well.

I agree with you. I’m a software engineer and over my 25-year career this is by far the biggest transformation in the field. It’s not so much how much time it saves but how it opens up possibilities. I had never done any GUI work but last Sunday I managed to write an application that was useful to me within a day. Without LLMs I wouldn’t have completed it due to the number of issues I had to solve. Now I just ask one LLM to solve my issues and if it fails I ask another one.

I read somewhere that even just a few years ago there were Getman business leaders and politicians who did not believe the Internet was important.

When I read Gary Marcus and the comments from most of his readers I have the feeling they’re missing the forest for the trees — yes LLMs make mistakes but regardless they’re very useful.

Thank you for your comment. I feel a bit more sane.

Expand full comment

100%%%%%

Wen, I calculated that the speed gain I has with AI coding is 1000:1

Never having used Python (but a lot of experience coding) I wrote a video editors in 2 days. I don’t use Adobe anymore.

I wrote a series of work tools for myself when a client asked a very hard question and I built a prototype in 4 hours. My team wanted to sell it for $2M

I have an illustration library manager which took me an hour, better than anything on the market hands down.

It’s crazy.

People keep imagining the tool must be autonomous, and they utterly miss the point.

Expand full comment

That wine glass is priceless. Did it mention if the glass was half empty or half full? I'd love to see what it produces with the phrase, "my cup runneth over." Can you run that one for me, Gary?

Expand full comment

I believe “My cup runneth LLMty” would be the applicable phrase in this case.

Expand full comment

Musk must have consulted Grok, as to whether he should have done a Nazi salute and it's impact on Tesla sales. Obviously it must have hallucinated. 😀

Expand full comment

I often find myself wondering what the AI cheerleaders online think will happen if AGI is ever achieved. Unless your net worth is at least in the high nine figures, you have exactly zero reasons to be excited about it.

Expand full comment

Obviously they plan to worship this mythical AGI as a god and surrender all their privacy, freedom and individuality to its will, since it will presumably either save the world or end it.

/s

Expand full comment

The cheerleaders just want to live in exciting times during which the world changes dramatically.

Expand full comment

That’s nice. We’ll see how excited they are when they’re with the rest of us plebians in the bread line hoping that almighty AI deems us worthy of sustainence.

Expand full comment
1dEdited

Many of them would literally prefer that to their current lives, because then they could be part of the epic story of #AIresistance. They're John Connor wannabes.

Happily, true AGI is at minimum decades away.

Expand full comment

Honestly I think they genuinely think they'll get a super nice UBI check and never have to work again. People who believe that are what we in the biz call "not very smart"

Expand full comment

Kinda ominous in a depressingly humorous way that WALL-E, as opposed to I Robot or Terminator, is the most accurate prediction of our future with technology thus far. That is, instead of being controlled by superintelligent robots and software, its more likely we'll have intelligent, but still very dumb robots running around doing near everything for us, while we increasingly become reduced to helpless cattle that rely on these bots to run our lives. Also, the Earth is screwed and getting wrecked too, that one always gets overlooked, but it's an equally ominous future for the world with its own significant consequences.

Expand full comment

Surely, the bots will save us from ourselves.

And from Shirley.

Expand full comment

Pro tip: You can tell a cow's religion by the language it speaks. That's how the rabbis tell.

If they hear Latin it's Roman Catholic, Presbyterians speak Gaelic, evangelicals speak in tongues... you get the idea. Only Hebrew-speaking cows can be slaughtered and prepared Kosher.

Doesn't everyone know that?

Expand full comment

I thought all cows were moo-slim.

Expand full comment

Bravo.

But are they halal or haram? I forget.

Expand full comment

Gary, will you be attending nextAI? ... I'd like to introduce you to our solution that addresses some of the problems that you express are not getting solved by the big frontier labs, but at the same time nobody is listening to smaller innovation plauyers that might have interesting solutions to offer

Expand full comment

It's going to implode when the models are trained on internet content that is based on the output from the previous generation of models.

That generation will deserve the name ouroboros.

Expand full comment

Do you realize that DALL-E 3 (OpenAI’s image generator) is totally separate from any GTP, and is now a year and a half old? The only thing GPT-4.5 is doing is parsing your prompt and feeding it to DALL-E, but it has no actual role in image generation.

Expand full comment

But if GPT 4.5 were actually as “intelligent” as it is touted to be, it seems like it would easily “see” the obvious problem with the generated wine glass image.

If bots like GPT 4.5 are acting as true “intelligent agents”, they should surely be able to assess the output of the modules like DALL-E that they are interfacing with and not simply return unphysical images and other nonsense.

Expand full comment

Anyone else see the WSJ article this week on one of the big banks, I believe JP Morgan, and genAI? The exec kept insisting on the value while quietly admitting that there were no measurable gains, in spite of an enormous user base….

Expand full comment

If you can dig this one out, it would be really appreciated. It might help with the hype over here.

Expand full comment

It’s called The Rise of Artificial Intelligence at JPMorgan, from the Feb 24 WSJ. I can’t share a link as I read it via an aggregator. When asked about tangible productivity gains, she said it was “early innings” and went on to hypothesize about future gains without evidence. She also declined to provide specific areas with good returns on AI and said all businesses 🤣😱

Expand full comment

That tripe bit is priceless.

The new Claude does seem good, for what it is.

Expand full comment