(no title)
jopsen | 9 days ago
Right now sota models requires a lot of iron.
It's possible that this will always be the case. But its is not a certainty!
We've seen software improvements shave orders of magnitude of compute requirements before. This could totally happen here. Iron could easily become stranded assets.
But that said, models have already become commodities, well somewhat. Is the value in running inference or applying it?
Today, we dare not use vibe coded libraries for mission critical things, HTML sanitization as an example.
But one day, who is to say the industry won't be disrupted by a vibe coded database with ~100% Oracle compatibility? Made by a nerd in a garage.
Established code bases is a moat today. It might not be in 5 years. Big tech won't be well positioned to take advantage, because trusting vibe coded crap is risky.
My point is mostly: the future is uncertain. Big established software companies might see their moat challenged by nerd in a garage running LLMs in the cloud.
What about the Adobe suite? AutoCAD? Office, etc. (To be fair, it's possible that software never was the moat).
zdragnar|9 days ago
This is the answer to all of your questions. Network effect and brand recognition sell Oracle, Adobe, office etc. Alternatives to all of them already exist, with either feature parity or close enough for most people.
The existing brands keep going because big companies and institutions don't pay for products vibe coded by some guy in a garage, they buy products that have paid support that they know will continue to exist for years.
jopsen|9 days ago
But what about 5 years from now?
What when the menus have the same layout, compatibility with the legacy binary file format is near perfection.
Today, alternatives exists, but they are not polished the same way.
bigstrat2003|9 days ago
Based on the abysmal ability of LLMs to write code today, that's not likely to happen. One never knows. But I wouldn't put money on it.