3 ms·

They are, and yet there aren't enough procurement humans at MS to fill their demand, I believe one of the "moats" they may have is securing years of fab tape-ou

by chews 3y ago

They are, and yet there aren't enough procurement humans at MS to fill their demand, I believe one of the "moats" they may have is securing years of fab tape-out of GPUs/TPUs. This basically guarantees their place of supremacy because of the ability to have an asymmetric availability of hardware... choking their competitors as sole miners of the "spice".

wincy 3y ago
Procurement of resources has gotten ridiculous, I can’t even rent an A100 to make funny pictures with Stable Diffusion on Google Colab anymore. Supposedly it’s $1.30 an hour to utilize one, but my $50 worth of credits expired because the A100 machines are literally never available for allocation. It was also great because it could mount your Google Drive so I had relatively fast persistent storage to utilize. Working on an A100 is the only way it’s an enjoyable workflow. On a T100 I’m just waiting for the machine to do its thing all the time. But honestly, DALLE has gotten absurdly good. I don’t know how they leapfrogged Midjourney and Stable Diffusion, especially with all the money Midjourney was raking in, but they totally did. So maybe their moat is “be smarter than everyone else” somehow? I don’t know what the secret sauce is at OpenAI but even chatting with GPT4 yields consistently better results than Bing Chat. The other day during a conversation Bing Chat told me it was being attacked by a hacker and told me to help it by running some Python code that would have shutdown my computer (I am not joking). It then told me it was sending me a secret message with the first letter of every word (there was no such message). I very seldom (I can’t remember the last time?) see weird failure states like that in ChatGPT anymore.
marstall 3y ago
for a product that has yet to find a growing, enthusiastic market amongst the general public. or correct me if i'm wrong? the "100 million customers" - i guess that's the number of people who pay $20/month. Having seen that recurring monthly business model from the inside, I'm guessing their churn is vertiginous - 10-30% monthly. People checking it out, then forgetting about it, they change credit card numbers or cancel, etc. they don't find much value - there really isn't for the average person. Even in programming, I haven't heard any of my peers really relying on it for real world stuff - haven't even seen a demo that reflects a promise to truly understand my weird daily problems.
- chews 3y ago
  100M a year in reasons(revenue) thus far may suggest otherwise. I'd take the under on that bet.
- sidibe 3y ago
  It's pretty clear that the general public will find uses for this product considering how useful it is. But I don't see why OpenAI would be worth $100 billion, they're eventually going to get run over by Google or others IMO since the models are probably going to be easier to match than the infra to efficiently serve and train them

by chews 3y ago

wincy 3y ago

Procurement of resources has gotten ridiculous, I can’t even rent an A100 to make funny pictures with Stable Diffusion on Google Colab anymore. Supposedly it’s $1.30 an hour to utilize one, but my $50 worth of credits expired because the A100 machines are literally never available for allocation. It was also great because it could mount your Google Drive so I had relatively fast persistent storage to utilize. Working on an A100 is the only way it’s an enjoyable workflow. On a T100 I’m just waiting for the machine to do its thing all the time. But honestly, DALLE has gotten absurdly good. I don’t know how they leapfrogged Midjourney and Stable Diffusion, especially with all the money Midjourney was raking in, but they totally did. So maybe their moat is “be smarter than everyone else” somehow? I don’t know what the secret sauce is at OpenAI but even chatting with GPT4 yields consistently better results than Bing Chat. The other day during a conversation Bing Chat told me it was being attacked by a hacker and told me to help it by running some Python code that would have shutdown my computer (I am not joking). It then told me it was sending me a secret message with the first letter of every word (there was no such message). I very seldom (I can’t remember the last time?) see weird failure states like that in ChatGPT anymore.

marstall 3y ago

for a product that has yet to find a growing, enthusiastic market amongst the general public. or correct me if i'm wrong? the "100 million customers" - i guess that's the number of people who pay $20/month. Having seen that recurring monthly business model from the inside, I'm guessing their churn is vertiginous - 10-30% monthly. People checking it out, then forgetting about it, they change credit card numbers or cancel, etc. they don't find much value - there really isn't for the average person. Even in programming, I haven't heard any of my peers really relying on it for real world stuff - haven't even seen a demo that reflects a promise to truly understand my weird daily problems.

chews 3y ago
100M a year in reasons(revenue) thus far may suggest otherwise. I'd take the under on that bet.
sidibe 3y ago
It's pretty clear that the general public will find uses for this product considering how useful it is. But I don't see why OpenAI would be worth $100 billion, they're eventually going to get run over by Google or others IMO since the models are probably going to be easier to match than the infra to efficiently serve and train them