50 Comments
Jul 24Liked by Gary Marcus

Driverless cars, Google glasses, Metaverse, and now AI. We just live in a world of hype unsupported by hard facts. Ultimately, the market will make the choices

Expand full comment

Don't forget crypto! I think investors are just really itchy for crazy good returns. The AI hype started right after the crypto market tumbled and people started to lose interest. Once the AI hype dies down, people will find something else to be super hyped about.

Expand full comment

Waymo is just great. Haven’t used Uber since Jan.

Expand full comment

I generally agree with lack of technical moat, and certainly with the profitability gap - we'll see if their investors are as patient as Amazon's were.

But there is also a brand recognition moat of "ChatGPT" as the default go-to. Easier for Google than for Anthropic to crack, via installed Search hegemony.

As to LlaMa 3.1 it is meant for connoisseurs so far, not for the general public. They are trying to be the Unix of the LLM world.

Expand full comment

Amazon had a point-at-able use case: mail order.

Amazon is basically Sears & Roebucks with an Internet catalog.

Expand full comment

I agree with your "general public" comment. unless someone comes up with a UI which abstracts the public from the API and internals of running Llama, there will be little interest from the public for its usage.

Expand full comment
author

about 10,000 startups are probably racing to do that tho

Expand full comment

I am sure

Expand full comment

For those of us who remember the dot com boom and bust know without a doubt that hype and no profit will eventually run out of steam. This is forward thought column by Gary that I feel must be really digested with clarity and discernment.

Expand full comment

Where is the application that will generate tens of billions of dollars to cover all their costs and those of Nvidia, and develop their own chipset, yesterday's news?

Expand full comment

Anyone bullish on Nvidia?

Expand full comment

Absolutely, whether the future is AI, VR, AR or some combination of them, massive processing power is needed and nVidia are leaders in that domain. That said, I'd be hesitant to buy in at current prices because they're priced to continously beat expectations for future quarters, which may be stymied by either general slowdown of capex expansion or an AI drought if LLMs turn out not to continue scaling. If they miss a single earnings the stock will plummet. I would 'buy the dip' if that happened, though

Expand full comment

Is OpenAI really paying its alleged 2500 staff $600k on average? I need a change of career.

Expand full comment

Average wage is a mostly meaningless measure. 2500×600−2499×250=875250. You can pay the king $0.875250G, the remaining 2499 staff $250K and arrive at $600K average.

Expand full comment

The minimum base salary is about 300k, so it's not surprising that average is 600k, given many multi-million earners.

Expand full comment

The problem for those of us who do AI research for a living is that they waste a generation’s worth of funding on dead ends and repeating each other’s work. We then get another 20 years of AI Winter as everyone nurses the disappointment and the financial loss. Believe me, we have been here before, more than once.

Expand full comment

ChatGPT's killer app is student cheating / plagiarism. Apart from that, generative chatbots are useful for generating bullshit and misinformation, especially on social media. In the business world, remember that marketing is all about telling stories. There are probably great ChatGPT applications for social media marketing and communications, which are a recognized and fertile ground for chatter, fiction, nonsense and sadly sometimes just bullshit.

Expand full comment

I don't think OpenAI needs to have a "killer app" of their own. They provide the platform (LLM) over which developers are supposed to build their own product –that is, their app.

Public interest in ChatGPT can decline over time (it has, as a matter of fact) but OpenAI money is coming from many small companies accessing its platform. At least this is the way I guess Altman is planning OpenAI's business.

Expand full comment

I think that ChatGPT has been baked into a lot more apps, and they can probably use that as a "moat" for a while. but APIs can be re-written.

Expand full comment

"Last Fall, I warned that they might someday be known as the WeWork of OpenAI" typo, it's "the WeWork of AI"

Expand full comment

I think the datacenter builders and cloud providers (oracle, etc.) using Nvidia equipment will definitely be the WeWork of AI. And unlike buildings, technology depreciates quickly. You can build a building with a 30 year bond. Datacenters, not so much.

Expand full comment

And it’s still a solution in search of a problem. Where is the killer use case? Everyone is spinning their wheels because of all the hype, but the promise remains vague.

Expand full comment

I dunno about you but Claude Sonnet has artifacts and excellent support, and it still seems to code better. Anthropic remains top of the game for me.

Expand full comment

"Among other things, I warned that “although the underlying technology initially improved rapidly, leading to a lot of excitement, it may soon, perhaps this year or next, reach a plateau”,

The underlying technology thinks it can get away without knowing the meaning of anything.

To handle meaning would require completely different people and a lot of work. It is not in the wheelhouse of OpenAI or Zuck or Google.

Expand full comment

Just wait until they lose The NY Times copyright lawsuit.Then their time will truly be up.

And Microsoft CEO Nadella will be wishing he’d never heard of OpenAI and Sam AI’m

Expand full comment

I do not like

To pay a dime

I do not like it

Sam AI’m

Expand full comment

With apologies to Dr. Seuss

Expand full comment

And everybody forgets about a model of human language - a Symbolic Language Model that is a foundation for education and science - an initial goal of Internet. OpenAI is not a "company" it is a knowledge base of humanity and is a tool for creating a model of language. OpenAI is unique in this role.

Expand full comment

Take a history of Uber as a project for investors and now for me Uber is a synonym for a taxi.

Expand full comment
founding

"Gary Marcus’s predictions over the last couple years have been astonishingly on target."

–Gary Marcus

Expand full comment

If it's so costly to run OpenAi and Anthropic, isn't it also costly to run Meta's AI division? I guess Meta can pull revenue from their ads business to cover costs, but that doesn't mean they've figured out how to run a foundational model profitably either. And as a publicly-traded company, there will be shareholder pressure to find ROI sooner.

Expand full comment