Loading…
20
Shoutout to the podcast guy who said 90% of AI startups use the same 5 open source models
I was listening to a small tech podcast on my drive home from the office in Sacramento last week and the host casually mentioned that something like 90% of AI startups are basically just wrapping the same handful of open source models with a nice interface. That number stuck with me way more than I expected, because I always assumed there was way more secret sauce happening behind the scenes. So I started reading up on it and yeah, a lot of the "innovation" is really just fine tuning and packaging, not building from scratch. Kind of makes me wonder where the actual new ideas are coming from. Has anyone here actually tried building something on top of these open models, and did it feel like real innovation or just fancy plumbing?
1 comments
Log in to join the discussion
Log In1 Comment
sanchez.ivan10h ago
Funny thing is those 5 models probably all trace back to the same 3 papers from like 2017. So really you got a bunch of startups fighting over who can put the prettiest chat box on the same math. I tried fine tuning a Llama model for a side project last year and honestly it felt less like building a rocket and more like changing the oil on someone else's car. The "secret sauce" is mostly just good marketing and a pricing page. Not saying there's zero new stuff happening but a lot of these founders are basically reselling the same brain with a different hat on.
6