Sam Hogan 🇺🇸
4.6K posts

Sam Hogan 🇺🇸
@samhogan
ceo @inference_net flywheels and feedback loops
San Francisco, CA Katılım Mayıs 2012
1.7K Takip Edilen23K Takipçiler
Sabitlenmiş Tweet

@creebeauvoir I sold mine when I moved to SF. A true cultural betrayal
English

@ad0rnai Lan everyone thinks your billboards are for @inference_net. I have dozens of people tell me this every month. please do not take them down. it would be very tough for us
English

If the weather were like this 9 months per year, SF would be the #1 American city and it’s not even close
English

@creatine_cycle girl with an android who is into weird chemicals? brother i have bad news for you
English

Unfortunately training by itself is not a defensible moat without being coupled with inference. The unit economics simply don't work out ie training costs, if you train a good enough model that people want to use, should be a rounding error compared to inference costs. In addition to that, you become this awkward middle man between the customer and the GPUs/inference provider the model runs on. Training will eventually become commoditised in a sense, and the training teams left standing will be the ones embedded in inference who are either very good at creating the abstractions to enable anyone to train, or extremely hands-on and dedicated to enterprise
illiquid@lefttailguy
So is DoorDash churning from @appliedcompute? I suspected that something like this would happen. You essentially work with companies like Applied Compute and Metis to see how strategic the specific intelligence they can help you build is, and if it's strategic enough, you of course want to in house the development that stuff. Citadel doesn't buy trading outcomes from other companies.
English




















