Y Combinator's Latest Batch Is 35% AI Startups

More than one-third of the famed startup accelerator Y Combinator's latest batch of companies are focused specifically on artificial intelligence. From a report: Y Combinator received a record 24,000 applications for its latest cohort, and accepted just under 1% of those, Garry Tan, president and chief executive officer of the accelerator, said Tuesday in an interview on Bloomberg Television. About 35% of the companies selected for the program are AI-focused, he said, and as many as half involve AI as a component of their business. "There's something very special happening here," Tan said about AI in San Francisco. "The smartest people in the world are sitting in those cafes having discussions. Not just about starting their companies, but also what is the cutting edge of what these AI models can do."

Such a quick turn!

[istartedi]

Crypto winter set in, then *boom* a full-blown AI Summer came out of nowhere. Now it looks like we're in Crypto Indian Summer along with AI Summer, so you know what this means: AI driven Crypto bubble.

You heard it here first.

The party isn't over until ChatGPT 7 fat-fingers Bank of America in to bankruptcy with one errant trade.

Re:

[CEC-P]

AI confidently told me that there will be no AI bubble and as we know, everything an AI says is true. Oh and "says" implies it knows what's it's saying and isn't mimicking human speech patterns of course. It definitely doesn't simply do that.

Re:

[ls671]

+1 came here to say basically the same thing. Back then, I am pretty sure about the same percentage of "Y Combinator's Latest Batch" were cryptos related

Re:

[HBI]

Everything essentially is a trial balloon pump and dump. If the VC wins on a small percentage of them, they make out. No one is about just like, running a business. That kind of thing is old fashioned.

Re:

[HiThere]

To be fair, most businesses are run pretty stably. You don't hear about those, though.

Re:

[istartedi]

I've always had the impression that failure was endemic, and the ever-turning store fronts in our locality would seem to confirm it but that's just anecdote. Here's some data [commerceinstitute.com], but confirming it is above my pay grade. A 50% five-year survival rate and a 30% 10 year survival rate doesn't seem that stable. OTOH, stable businesses might employ more people and generate more sales, so if you look at it from that PoV you might be able to say, "Most business is conducted by stable companies" and be confidently c

Re:

[HiThere]

IIUC, there's a high initial failure rate, and then they run stably until...
Unfortunately, there's lots of untils. Management changes is one. Sometimes the business environment changes. And sometimes it's just the owner gets tired of running the business. So, yeah, there are lots of crisis situations, and there's probably some kind of exponential rule. I wasn't really thinking about large businesses, though. Still, "Levi's" has been in business a long time, and you can point to several others. Tech c

get ready for the bubble to pop

[CEC-P]

Hey, remember the big investment push and bubble for web services and web-based companies and ecommerce? At least those websites worked and at least there was a provable market for computing and online services, etc. This time everyone's all excited about an "AI" that has absolutely no idea what it's saying, mimics speech pretty well to fool humans, but can't be used in a lawyer's office because it will simply lie to you and make things up that didn't happen. You definitely can't use it in a medical field.

Re:

[gtall]

I see no reason specific AI bots cannot be trained on good data for a narrow field and be quite useful. They don't all have to use LLMs.

Re:

[Visarga]

You are not wrong about AI being untrustworthy. But it's a tool - you use it where it makes sense. It's also a very complex tool that can be "prompted" or "fine-tuned" to reduce its bloopers. The rate of advancement is amazing, but that doesn't mean it will become reliably autonomous anytime soon.

Re:

[HiThere]

What's "soon"? Within narrow domains it's already more accurate than the average person. And where you don't need it's capabilities, it can use standard computer tools to produce reliable performance.

ChatGPT was developed as a research project, and wasn't optimized to be reliable. (OTOH, I think that optimizing it for reliability would have severely reduced its capabilities. Human language isn't that precise, after all.)

FWIW, I expect the CURRENT generation of LLMs to replace (parts of) LOTS of jobs. T

what!? no blockchain?

[Anonymouse Cowtard]

It's not AI, it's G-AI. In 5 years we'll all be asking wtf it was all about. It's a fad. It has a use case but not the be-all-end-all. It's like a reverse history eraser in Photoshop. It's like a parrot swallowed a few reference books and all the biased comments made online in the past few years. Go ahead, throw your money in. What would I know?

Re:

[Linux Torvalds]

Translation: "This 8-bit microcomputer is useless. It can't even run Crysis. I don't get the hype."

Maybe wait for the second-generation models to come out before drawing such sweeping conclusions.

Re:

[narcc]

8-bit microcomputers had immediate practical applications. As for AI, the "second-generation" has long-past come and gone.

As predicted, reality has tempered expectations: "Okay, maybe this time it won't do all those things I said, but just you wait until the next generation. Then you'll see!"

AI: just 10 years out since 1956!

get rich quick

[peterww]

If any of ya'll are trying to fill your pockets with millions of VC money, just come up with some bullshit with "AI" in the name. It's a great time to fleece some greedy stupid businessmen!

What hubris

[GlobalEcho]

Garry Tan is quoted as saying

The smartest people in the world are sitting in those cafes having discussions...

What hubris. Only somebody from (and partisan to) the Bay Area, or maybe Harvard, would say something like that.