[ericmcer]

Seriously you can ship in a week things that FAANGs would have paid billions for 10 years ago.

LLMs are just glue between pieces of your code you still need to be able to plug them into a coherent architecture to do something impressive.

[altmanaltman]

Apple Pay (launched around 12 years ago, and likely costing billions) is exactly the counter-example here. The 'code' was the easy part. The moat was the decade of hardware R&D and the leverage to force banks to adopt a new standard. An LLM might write the API wrapper in seconds, but it can't hallucinate a relationship with Visa and Mastercard. You literally cannot create a new Apple Pay in a week or even years, no matter how much you vibe code.

I'm sometimes baffled by what people think can pass as a product in a real sense.

[sarchertech]

To the extent that’s true it has much, much more to do with AWS, open source libraries, and collective knowledge, than it does LLMs.

But I honestly can’t think of anything you could do in a week that a company in 2015 would have paid billions for unless it’s something like tweaking an LLM. But in that case it’s the original model, not the 1 week or work you put in.

[tdrz]

OK, say you build a Whatsapp clone in a week. How many Whatsapp users will switch to your app?

[zabzonk]

> things that FAANGs would have paid billions for 10 years ago

such as?

[lich_king]

I can only assume the grandparent means Google+

[ericmcer]

I mean literally anything that leverages modern APIs.

WYSIWG Site Builders, text Chat bots, audio Transcription, voice synthesis.

Yeah building from scratch would take longer, but you can slap a UI, a DB/schema around modern APIs and output something that would be science fiction 10 years ago.